Frontpage posts - Effective Altruism forum viewer

Marginal Victories: career advising and opportunities for U.S. democracy preservation & political work by Annika Burman 🔸

Annika Burman 🔸 — Mon, 20 Jul 2026 14:33:48 +0000

TL;DR: Marginal Victories is a new initiative to provide 1:1 career advising, opportunities, and resources for people exploring high-leverage U.S. democracy preservation and political work. Built by impact-oriented people doing pro-democracy work, the early MVP is now up at marginalvictories.org. Fill out the 10-minute form now to receive these resources as they become available over the next few weeks.

Why this should exist

Over the past year, through 1:1 conversations I’ve had with people working on U.S. democracy, I’ve grown increasingly convinced that this work is tractable. If I hadn’t had these conversations, I doubt I would have come to the same conclusion — most organizations in this space don’t post about their work online due to its sensitive nature, and as a result, it’s hard to find examples of people breaking into this space and quickly getting results, even though there are plenty of success cases.

Our goal with Marginal Victories is to create a pathway for people considering transitioning to work on U.S. democracy and politics to get plugged into this field and hear about high-leverage opportunities they can get involved with. If all goes well, we will have more talented, impact-minded people working on this cause area.

Who is this for

You’ll probably get value from this if you’re:

Considering whether or not you should start a career, or transition into, working on U.S. democracy preservation/politics
Actively looking for roles in this space, especially if you don’t know what orgs are doing impactful work
Looking to start a new project relating to U.S. democracy or politics — we have a database of expert-generated ideas and can potentially help you get resourced

What resources will be available and when

Advising

The primary offering of Marginal Victories. We expect actual advising calls to start within the next month, but you can register your interest when you sign up today.

If you’re interested in becoming an advisor (i.e., taking a few advising calls per month/week, possibly for an hourly rate), please fill out this form.

Other resources

Job/opportunities board—ready now! Roles relevant to your career stage and skills will be sent to your inbox monthly.
Cause prioritization arguments—ready to share next week; if you have remaining uncertainties, you can express interest in an advising call for this purpose.
- If you have additional recommendations for what resources to share, submit them to this feedback form.
Ideas for new projects—If you think this might be a good fit for you, we recommend expressing interest in an advising call. We have existing ideas for projects that should exist, but we’re also open to hearing yours.

In the future, we may also set up the options to access donor advising, recommended resources for building context, networking, volunteering opportunities, and more. Open to suggestions!

What we’d love from you

Fill out the form to access the resources: marginalvictories.org
Let me know if you encounter any issues via the feedback form.
If you’re working full-time in U.S. democracy preservation/politics and you’re interested in serving as an advisor for even 1 hour/month, please fill out this form.

Thanks!

Annika + Marginal Victories Team

Deconstructing the AI Ecosystem: How Do We Isolate Component CEA within a Multi-Modal Child Wellbeing App? by Wongani Zulu

Wongani Zulu — Mon, 20 Jul 2026 14:25:50 +0000

Deconstructing the AI Ecosystem: How Do We Isolate Component CEA within a Multi-Modal Child Wellbeing App?

Executive Summary

I recently won funding to develop a multi-modal, AI-enabled application for Community Health Workers (CHWs) in rural Zambia. The app captures voice narrations in local languages and uses computer vision to assist with diagnosis and multi-sectoral triage across Health, Nutrition, and Child Protection.

While traditional NGOs love integrated programming, it presents a classic challenge for EA-style scrutiny: causal dilution.

I am new to the EA community. My background is in ICT and monitoring, evaluation, research, and learning (MERL) in Zambia, rather than pure academic economics. I am writing this post to introduce myself and to solicit feedback on how to mathematically disaggregate and evaluate the cost-effectiveness of a single automated path within a complex, multi-sectoral digital ecosystem.

The Context: Field ICT/MERL Background Challenged by EA Rigor

By way of introduction, I manage MERL for a portfolio of 33 grants worth $25.7M at World Vision Zambia. My career has centered on building data infrastructure to track massive interventions across health, nutrition, education, social protection and child protection, including:

Zambia Multiple Micronutrient Supplementation (ZMMS): Building the data ecosystem for the ongoing implementation research project.
World Food Programme (WFP): Developed a management information system tracking all strategic outcomes and a mobile app for tracking beneficiaries of COVID-response social cash transfers.
Zambia National Malaria Elimination Programme (NMEP): Developed tracking systems for the ITN mass distribution campaigns, alongside an app monitoring the distribution of Against Malaria Foundation (AMF)-supplied ITNs across six provinces.

While I have co-authored a paper on WASH (2025) and a malaria poster for ASTMH (2022), because of my interest in environmental science, my core skillset has generally centered on data and database management, field logistics, and grant compliance.

However, the little I’ve learnt about EA so far has challenged my thinking. Traditional development focuses heavily on coverage, integrated interventions, and sustainability. But I have recently learnt that EA focuses heavily on marginal cost-effectiveness and unbundled, single-variable impact. I would like to deepen skills in bridging that gap using my new project as a case study.

How the app works: A multi-modal AI core

To mitigate low literacy and take the guesswork out of CHW diagnostics, the app processes unstructured, multi-modal data in real-time at the edge:

Audio narration and NLP: The CHW narrates the interaction in their local language (e.g., “I am with Betty, a 2-year-old girl who has been vomiting since yesterday… she has a bruise on her left elbow and her feet seem swollen”). The app uses Natural Language Processing to translate, parse keywords, and prompt the CHW to run specific tests (like an RDT for malaria or checking for pitting edema).
Computer vision and clinical automation: The app uses the phone camera to interpret Mid-Upper Arm Circumference (MUAC) readings, analyze video for chest in-drawing or fast-breathing (pneumonia tracking), verify pitting edema via the 3-second rule (SAM detection), and capture patterned bruises to determine if a child suffered physical abuse.
Dual-layer logic: It generates a baseline opinion using hardcoded clinical parameters, paired with a second opinion driven by Retrieval-Augmented Generation (RAG) mapped directly to standard WHO and Zambian Ministry of Health treatment guidelines.
Automated multi-sectoral triage:
- Health: Severe malaria (and other cases that CHWs are not allowed to treat such as pregnant women) alerts instantly update health facility databases so clinic staff are prepared for the incoming referral.
- Nutrition: Growth-faltering triggers immediate, localized nutrition protocols.
- Protection: Suspected physical abuse text alerts are sent anonymously to Social Welfare and the record of the suspected case is erased from the CHW’s device.
- WASH: An algorithmic trigger flags clusters (e.g., if 5 cases of diarrhea occur within 5 days in one locality) and alerts WASH teams to potential water-point contamination.

To operate seamlessly in far-flung, off-grid Zambian villages, the app is built as an offline-first system. Core clinical logic and WHO/ Ministry of Health diagnostic parameters are entirely hardcoded on the device to prevent latency. The system queues unstructured audio and multimodal inputs locally, executing vector store queries (via Vertex AI) for complex edge cases, and pushing batched data packages to the centralized dashboard only when network connectivity is available.

Technical Puzzles I Need Help Solving

Because the app dynamically triggers distinct pathways from a single audio string, calculating standard Cost-Effectiveness Analyses becomes highly complex. I would love the forum’s thoughts on three specific methodological hurdles:

Puzzle A: Shared Cost Allocation in Multi-Modal Architecture

The cost of building the core LLM pipelines, audio transcription models, and RAG architecture is shared equally across every module.

The Question: When calculating a CEA specifically for the Malaria referral component or the Nutrition component, how should these fixed technical overhead costs be apportioned? Should they be split strictly by the volume of algorithmic triggers tripped (e.g., if 60% of prompts end up being malaria-related, does malaria absorb 60% of development costs)? What is the EA standard for cost-accounting in shared AI infrastructure?

Puzzle B: Causal Attribution of Automated Cross-Pollination

Let’s say a CHW opens the app strictly because of a child protection concern, like a bruise. During the voice narration, the AI picks up on an auxiliary symptom or the camera catches a low MUAC reading. The app prompts a nutrition intervention, and the child’s health status subsequently improves.

The Question: The user’s primary intent was protection, but the health outcome was nutritional. How can we rigorously structure telemetry data to assign a causal weight to an outcome when an AI core actively forces cross-pollination across distinct sectors?

Puzzle C: Commensurability (Child Protection/WASH vs. DALYs)

Quantifying the cost-effectiveness of malaria interventions or micronutrients may be straightforward because we can map them directly to DALYs or mortality reductions. On the other hand, to me it seems automated social welfare alerts or automated WASH cluster triggers are kind of trickier to convert into standard EA units of impact.

The Question: How do EA researchers handle multi-sectoral programs where one arm produces clear DALY metrics but another arm produces preventive protection or clean-water system metrics? Do you try to convert structural alerts into DALY equivalents, or do you evaluate them on entirely separate tracks?

Proposed Evaluation Ideas (Please feel free to critique)

To isolate impact for future scaling, here are a few ideas I am considering to ground-truth the data. Please tear them apart:

A/B Feature Gating: We randomly assign rural zones into different cohorts. Cohort 1 gets the app with only the core health/malaria pathways active. Cohort 2 gets the fully integrated health, nutrition, WASH, and protection pipelines. We then measure the marginal delta in specific health outcomes between the cohorts over 12 months. Is this logistically ethical and statistically viable for an early-stage digital tool?
Counterfactual Mapping via NMEP/Government Baselines: Utilizing my background with the Zambia National Malaria Elimination Programme data ecosystems, we map our app’s specific malaria and pneumonia referral success rates against neighboring control districts that rely purely on traditional, non-digital paper triage.

Moving Forward

I am deeply fascinated by the level of rigor the EA community brings to global health. While my day job involves managing large, complex portfolios where integrated programming is the administrative norm, I want to train myself to think at the marginal, unbundled level.

I would highly appreciate any reading recommendations, existing CEA frameworks for multi-sectoral AI products, or blunt critiques of my proposed methodologies.

Note: If you want to look at some of my past field data infrastructure work, I have linked my co-authored 2025 WASH paper and 2022 ASTMH malaria poster on my personal website: wong-zulu.guru/portfolio.

After Meat: What Happens to Animals When We No Longer Need Them? by RufaroNyakudya

RufaroNyakudya — Mon, 20 Jul 2026 14:22:41 +0000

Electric sheep, obsolete animals, and the moral danger of valuing life only through utility

By Rufaro Daniel Nyakudya

Imagine that the last industrial slaughterhouse does not close because of a protest, a new law, or a sudden wave of human compassion.

It closes because it is no longer competitive.

The supermarket shelves remain full. Burgers still sizzle on grills. Chicken remains part of family meals. Milk still softens coffee, leather-like materials still cover shoes and car seats, and pharmaceutical companies still manufacture products once derived from animal bodies. The difference is that these goods now come from cultivated cells, plants, fermentation tanks, engineered microorganisms, and synthetic biology.

The food has not disappeared. The animals have.

Not all at once, of course. No government opens the gates of factory farms and releases billions of chickens, pigs, and cattle into forests and cities. Instead, farmers breed fewer animals each year. Herds shrink. Hatcheries close. Barns stand empty. The remaining animals age, die, and are not replaced.

Within a few generations, animals once visible across vast agricultural systems could become uncommon.

This is normally presented as a story about the future of food. But the more difficult question is not what humans will eat. It is what happens to our relationship with animals once we no longer need their bodies.

Electric sheep in reverse

Philip K. Dick’s Do Androids Dream of Electric Sheep? imagines a world devastated by ecological catastrophe. Many living animals have become rare or extinct. To own a real animal is to possess something precious: a sign of social standing, moral respectability, and perhaps continued connection to the living world. Those who cannot afford real animals may keep convincing electric substitutes.

Rick Deckard, the novel’s protagonist, owns an electric sheep and longs to replace it with a living animal. His society claims to value empathy, but it also separates beings into categories: those who may be protected, those who may be owned, and those who may be “retired.”

Animals and androids are not merely background details. They expose the instability of the boundaries humans use to decide who matters. Literary scholar Sherryl Vint argues that the novel’s ethical concerns cannot be understood by focusing only on the androids and ignoring the animals. The species boundary is itself political: it helps societies decide whose experiences count and whose exclusion can be justified.

A post-meat society would reverse Dick’s premise.

In his world, technology imitates animals because living animals have become scarce. In ours, technology may imitate the products made from animals so successfully that the animals bred to produce them become scarce.

We might have meat without slaughter, milk without cows, leather without skin, and medicine without animal experimentation. Yet the same technological success that removes animals from production may also remove them from everyday human life.

This could create a strange moral test. Will animals become valuable because they are finally free from economic exploitation? Or will they become irrelevant because the products we wanted from them are now available elsewhere?

The farmed-animal population paradox

One of the most common reactions to the idea of a post-meat society is practical: what would happen to all the farm animals?

The question often imagines an enormous, simultaneous release. Billions of animals suddenly wandering through fields, roads, and forests, unable to feed themselves and threatening local ecosystems.

That is almost certainly not how a transition would happen.

The vast populations of chickens, pigs, cattle, sheep, and farmed fish alive today are not stable populations existing independently of humans. They are continuously reproduced because markets demand their meat, milk, eggs, skin, and other products. When demand falls, farmers do not normally release animals. They breed fewer of them.

The liberation of farmed animals might therefore look, numerically, like their disappearance.

This creates an uncomfortable paradox. Ending factory farming would prevent future animals from being born into systems structured around confinement, rapid growth, and early death. But it would also mean that many domesticated breeds would exist in far smaller numbers, and some might disappear altogether.

Would that be a tragedy?

There is no single answer. Some livestock breeds carry cultural histories, locally adapted knowledge, and forms of genetic diversity developed over generations. Their disappearance could therefore represent a genuine loss to human communities, especially where they are tied to identity, heritage, climate resilience, or traditional ways of life.

But what counts as loss depends on whose perspective we consider. A breed may be valuable to humans while the animals within it continue to live under conditions of confinement, extraction, or early death. From their perspective, the end of such a system may look less like extinction and more like liberation from being bred for human use.

This tension is especially sharp for commercial animals selectively bred so intensively that their bodies themselves generate suffering. Chickens bred for extreme growth can experience mobility and cardiovascular problems, while dairy cattle bred for very high milk production may face serious metabolic and reproductive strain.

The now-extinct turnspit dog offers a telling precedent. Bred to run inside wheels that rotated roasting spits, it disappeared when technology made its labour unnecessary. Its loss erased part of human cultural history, but few would argue that justice required continuing to breed dogs for the wheel. The ethical question is therefore not simply whether a breed should survive, but what kind of lives its continued existence would involve. We may want to preserve cultural and genetic diversity without preserving harmful biological designs or the systems of exploitation that created them.

Market forces should not make that decision alone. Breeds with distinctive genetic or cultural value might be preserved through gene banks or small welfare-centred populations, but only where doing so does not perpetuate suffering built into their bodies or the systems that created them.

Some animals could live in sanctuaries. Others might remain in carefully managed heritage populations. Certain domesticated animals could continue living alongside people without being bred for slaughter. But many would require permanent human care. They could not simply be “returned” to a natural state that no longer exists for them.

Even sanctuaries raise difficult questions. Who pays for decades of food, veterinary treatment, and land? Would the public fund animals it no longer eats, uses, or regards as companions? If private charity falls short, should governments redirect livestock subsidies, or should industries that profited from breeding these animals bear the cost? How many animals can realistically be accommodated? Should they be allowed to reproduce? Are they living for themselves, or have they become educational exhibits through which humans perform their new morality?

There may be no way to undo domestication completely. Humans created animals who depend on us, and that dependence creates responsibilities that do not disappear when the animals cease to be profitable. Perhaps our final responsibility is not to preserve every breed, but to care for the last animals well enough that no more need to be born into a dependence we created.

Animals without jobs

When imagining this future, we often ask: what role would animals have in society?

The wording reveals how deeply utility structures our thinking.

For most of human history, animals have been defined by what they provide. They are livestock, working animals, laboratory animals, guard dogs, dairy cattle, laying hens, wool-bearing sheep, breeding stock, or companions. Even conservation often describes wildlife through the ecosystem services it provides.

An animal without a function can appear almost conceptually incomplete.

But perhaps animals would not need a new role.

The most radical feature of a post-meat society may not be meat without animals, but animals without jobs.

Animals could exist without having to produce food, labour, materials, entertainment, or profit. They could form relationships, explore their environments, avoid pain, seek comfort, and pursue the forms of life available to them.

This does not mean humans would have no relationship with animals. It might mean relationships closer to those we have with wild populations: based less on ownership and use, and more on coexistence, habitat protection, restraint, and responsibility for the harms we cause. Yet the shrinking place of wild animals in a human-dominated world offers a warning. Animals without obvious utility may be valued in principle while being given very little space to survive in practice.

Former farmed animals could live in sanctuaries designed around their welfare rather than their usefulness to visitors. Humans might share landscapes with animals through forms of multispecies community and responsible stewardship. Some domesticated animals might remain companions, but this would force a distinction between rescuing animals who already exist and deliberately breeding new ones. If breeding ended, even dogs and cats might eventually become rare or disappear. Would that be a loss we should resist, or the gradual end of another relationship built on ownership and reproductive control?

Land released from intensive livestock production could be restored as wildlife habitat, forest, wetland, or grassland, allowing some wild populations to recover. But rewilding would require humility. Removing farms would not guarantee that land is used justly or that animals benefit.

Corporations could buy the land, enclose it for carbon-credit projects, convert it into luxury developments, or use it for intensive crop production. Conservation programmes could also displace communities that have lived on and managed those landscapes for generations.

Preventing this would require more than hope: secure communal tenure, participatory land-use planning, restoration conditions attached to transition funding, and public or community control of strategically important land.

Ending animal agriculture would create possibilities. It would not determine which possibilities society chooses.

A post-meat world may still be unequal.

It is easy to imagine the transition as a clean technological substitution: bioreactors replace barns, cultivated tissue replaces slaughter, and everyone moves smoothly into a more ethical food system.

But technologies arrive through political and economic systems.

Cultivated meat remains an emerging technology. Producing animal tissue without raising and slaughtering whole animals is technically possible, but large-scale production still faces questions involving cost, energy, growth media, bioreactor design, food safety, regulation, and consumer acceptance. Precision fermentation is already used to produce particular proteins and ingredients, but expanding it into a major component of the food system would require substantial infrastructure and investment.

A future in which alternative proteins dominate is possible, not inevitable.

No reliable timetable exists for cost parity, and technologies that become affordable in wealthier markets may remain inaccessible elsewhere for years.

And even if these technologies succeed, ownership will matter.

A small group of companies could control cell lines, genetic libraries, bioreactors, production software, and the intellectual property required to manufacture food. Countries without advanced biotechnology infrastructure might exchange dependence on imported animal products for dependence on patented food platforms.

The post-meat economy could become less violent toward animals while becoming more concentrated and unequal among humans.

Conventional meat might also survive as a luxury product. Instead of disappearing, slaughtered meat could become rare, expensive, and prestigious. We already see this logic in the appeal of artisan, organic, and “traditional” foods, where older production methods are marketed as more authentic precisely because they are less common. The renewed fascination with raw milk reflects the same romanticisation. In a post-meat society, the wealthy might consume slaughtered meat not from necessity, but as an expression of authenticity, tradition, or power.

Living animals themselves could become status symbols.

In a world where most people encounter animals through digital simulations, museums, or public sanctuaries, owning a living horse, cow, or rare breed could become the ultimate demonstration of wealth. Like the coveted real animals in Dick’s novel, their rarity could increase their social value without necessarily increasing respect for their autonomy.

We might stop treating animals as ordinary commodities only to reinvent them as luxury commodities.

Lacoste’s Save Our Species campaign offers a revealing, if well-intentioned, example of how rarity can acquire symbolic value. The company temporarily replaced its crocodile logo with ten threatened species and limited the number of shirts bearing each design to the estimated number of those animals remaining in the wild. The campaign used exclusivity to make extinction visible and support conservation. Yet the same logic could be inverted. In a post-meat world, living animals might become desirable precisely because they are rare: not beings to coexist with, but scarce objects through which wealth, taste, or moral concern can be displayed. Rarity can inspire protection, but it can also invite possession.

There is also a quieter dystopia: a society that causes far less animal suffering but becomes emotionally disconnected from living beings. Children might know cattle primarily as the historical source of a food now produced in steel tanks. Animals could survive mainly in gene banks, controlled reserves, zoos, private estates, and immersive simulations. If direct encounters with animals help people care about them, a greater distance could weaken the very concern needed to protect those who remain.

The violence would be reduced. But the underlying belief that nature matters only when preserved, consumed, or displayed for humans could remain intact.

A society can end slaughter without ending anthropocentrism.

There will not be one global transition

The phrase “animal agriculture” can conceal radically different realities.

An industrial chicken complex producing hundreds of thousands of birds is not the same institution as a smallholder household keeping goats, chickens, or cattle. A feedlot is not the same as a pastoralist community whose social, economic, and cultural systems are organised around livestock.

This distinction matters especially in Africa and other parts of the Global South.

For many rural households, animals are not simply units of meat production. They provide food, income, manure, transport, and draught power. They may function as savings and insurance where formal banking and social protection are limited. Livestock can also carry ceremonial, familial, and social meaning through bridewealth, inheritance, identity, and relationships between generations.

A technology that replaces meat does not automatically replace these functions. A pastoralist cannot simply exchange cattle for cultivated meat and be left economically whole. A smallholder may still need manure for crops, power for cultivation, an asset that can be sold during an emergency, and a source of household resilience. Removing livestock without replacing these systems could deepen vulnerability.

The transition must therefore be differentiated.

The strongest case for rapid change concerns large industrial systems that breed animals in enormous numbers and organise their lives almost entirely around maximum output. But even there, workers and farmers cannot be treated as collateral damage. Slaughterhouse workers, transporters, feed producers, veterinarians, processors, and rural communities may all depend on the existing system.

A just transition would require more than telling people to retrain. It could include debt relief for farmers with stranded infrastructure, income protection, locally appropriate education, investment in new food industries, and support for land restoration. The costs would be too high for charity alone. Governments could redirect livestock subsidies and other agricultural support, while development banks, climate funds, and philanthropic foundations could help finance infrastructure, social protection, and ecosystem recovery.

Governments could also establish time-limited food-transition funds financed through redirected subsidies, levies on large companies that profited from animal agriculture, and, once the new industry had matured, modest contributions from major alternative-protein producers. Imported and domestically produced alternatives could contribute on equivalent terms, although such charges would need to avoid raising the price of basic food or slowing adoption in lower-income countries. The revenue could be divided between support for displaced workers and communities and the long-term care of animals left behind by the transition. Any such fund would need a clear end date, independent oversight, and public reporting to prevent a temporary measure from becoming a permanent tax or a subsidy for politically connected companies.

NGOs could play a catalytic role by funding pilot conversions, providing technical and legal support, organising cooperatives, connecting producers to new markets, and helping communities design alternatives. Public and development finance would need to carry the highest costs, but NGOs could help turn that funding into transitions that work on the ground. Farmers should have opportunities to become owners and producers within the new food economy, rather than merely consumers of technologies created elsewhere.

In pastoral and smallholder settings, change would need to be co-designed with communities. It might involve gradually replacing particular animal products while expanding financial services, social insurance, alternative fertilizers, transport, energy systems, and other sources of resilience. Some relationships with livestock might persist where practical alternatives do not yet exist, even as communities seek to reduce the suffering and control involved.

The future of food should not become another project in which wealthy societies develop a technology and prescribe a universal pathway for everyone else.

There may be many post-meat transitions, moving at different speeds and reflecting different relationships between humans, animals, land, and livelihoods.

What we would owe the final generations

Suppose the technological transition succeeds. Demand for conventional animal products begins to fall rapidly. What would justice require?

The most dangerous period may be the gap between falling demand and a fully scaled alternative food system. Without managed breeding reductions, bridge finance, and guaranteed care, farmers could respond through early slaughter, neglect, or abandonment while consumers face shortages and price shocks.

The first obligation would be to prevent animals from carrying the costs of the transition.

Farmers facing collapsing markets might otherwise abandon animals, reduce veterinary care, or send them to slaughter sooner. Governments and companies that benefited from animal agriculture would need to fund the care of remaining populations and support gradual reductions in breeding.

The most credible model would spread the cost across redirected subsidies, a temporary industry levy, public transition funds, and charitable support rather than leaving individual farmers or sanctuaries to absorb it.

Publicly supported sanctuaries and retirement programmes could care for some of the final generations. Others could remain on transitioned farms under strict welfare protections. Those protections would require independent, risk-based inspection, animal-based welfare measures, traceability, public reporting, and penalties strong enough to make non-compliance more costly than compliance. Decisions about reproduction would need to balance the animals’ social and behavioural lives with the need to avoid recreating large dependent populations without adequate care.

The second obligation would be to the affected people.

Workers should receive income support, retraining, and meaningful opportunities in the emerging food system. Farmers should not be left with debt attached to barns, equipment, and herds whose value has collapsed. Communities should help decide how former agricultural land is used.

The third obligation would be institutional.

Governments might need to prevent conventional meat from becoming an opaque luxury industry with weak welfare protections. This could require regulating not only production, but also the breeding, ownership, sale, and consumption of animals and high-value animal products. Licensing, traceability, welfare standards, or outright restrictions might follow, but these could also create black markets or make animal use an exclusive privilege of the wealthy. Alternative-protein infrastructure should not be controlled entirely by private monopolies. Public research, open standards, and accessible licensing could help ensure that food security is not dependent on a handful of technology firms.

Animals themselves would also need stronger legal recognition. If their market value approaches zero, protections based mainly on ownership and commercial value may become even less adequate. Legal systems would have to recognise that animals have interests independent of whether anyone can profit from them.

A humane transition would not discard animals once they became obsolete. It would treat their dependence on us as a responsibility created by our own history.

The moral mirror

A post-meat future would force humanity to confront a question that industrial agriculture has allowed us to avoid.

Did animals matter to us, or did the things we extracted from them matter?

We have often valued animals according to their utility, intelligence, similarity to humans, emotional appeal, or economic contribution. Dogs receive one form of protection, pigs another, and rats another, even when the moral distinctions between their capacities are much less clear than our social rules suggest.

Technology might reveal this hierarchy by stripping away usefulness, even the quieter usefulness of companionship, affection, status, or emotional comfort. If animals are valued only for what humans can extract from them, no category is secure once that value disappears.

Once an animal produces nothing that cannot be obtained more cheaply from a machine, will we still protect it? Will we make space for beings who consume resources but generate no economic return? Will we care for lives that no longer serve us?

These questions extend beyond animals.

There is growing philosophical debate about whether future artificial systems could possess consciousness, agency, interests, or moral status. It remains deeply uncertain whether digital minds will ever exist in a morally relevant sense. We should not casually assume that current AI systems feel, suffer, or possess inner lives.

But advanced digital systems could still become more capable than humans in many economically important domains. They may perform much of the labour, analysis, and coordination on which society depends. They may become embedded in institutions that increasingly evaluate people through efficiency, productivity, risk, and measurable contribution.

The analogy is not that AI will inevitably treat humans as humans have treated animals. History does not function as cosmic punishment, and advanced intelligence does not automatically produce cruelty.

The comparison is a moral mirror.

We are building technologies inside societies that already rank life according to usefulness. We optimise systems for output, speed, and cost, then hope that these systems will somehow preserve values that our own institutions repeatedly fail to honour.

If we teach our most powerful technologies that the inefficient may be discarded, that the dependent have value only through what they produce, and that intelligence determines moral worth, we should not be surprised if those principles are eventually applied to us.

The future of food is therefore not only about what replaces meat. It is about what replaces the relationship that turned living beings into units of production in the first place.

A post-meat society could represent one of humanity’s greatest moral achievements. But only if animals are not discarded once they cease to be profitable, preserved merely as luxury objects, or remembered only through the products technology learned to imitate. The deeper transformation would be learning that a life does not have to be economically necessary to matter.

Perhaps the deepest question is not only what humans will do when animals are no longer useful to us, but what digital minds may do when human beings are no longer useful to them.

References

Dick, P. K. (1968). Do Androids Dream of Electric Sheep? Doubleday.

FAO. (2012). Livestock and Livelihoods. In World Livestock 2011: Livestock in Food Security. Food and Agriculture Organization of the United Nations.

FAO and WHO. (2023). Food Safety Aspects of Cell-Based Food. Food and Agriculture Organization of the United Nations and World Health Organization.

Heise, U. K. (2009). The Android and the Animal. PMLA, 124(2), 503–510.

IPBES. (2019). Global Assessment Report on Biodiversity and Ecosystem Services: Summary for Policymakers. Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services.

IUCN. (2019). IUCN and Lacoste team up again to inspire support for threatened species.

Long, R., Sebo, J., Butlin, P., Finlinson, K., Fish, K., Harding, J., Pfau, J., Sims, T., Birch, J., and Chalmers, D. (2024). Taking AI Welfare Seriously. Eleos AI Research.

Mogensen, A. L. (2024). Digital Minds II: Ethical Issues. Global Priorities Institute, University of Oxford.

Morais-da-Silva, R. L., Reis, G. G., Sanctorum, H., and Molento, C. F. M. (2022). The social impacts of a transition from conventional to cultivated and plant-based meats: Evidence from Brazil. Food Policy, 111, 102337.

National Trust Collections. (n.d.). Dog wheel (Object 149377).

Pottle, S. (2013). Segregating the Chickenheads: Philip K. Dick’s Do Androids Dream of Electric Sheep? and the Post/humanism of the American Eugenics Movement. Disability Studies Quarterly, 33(1).

Vint, S. (2007). Speciesism and Species Being in Do Androids Dream of Electric Sheep? Mosaic: An Interdisciplinary Critical Journal, 40(1), 111–126.

Engineering, AI Safety, and a Small Town in India: Thinking Out Loud by SaumyaSharma

SaumyaSharma — Mon, 20 Jul 2026 14:10:39 +0000

Epistemic status: exploratory. I’m genuinely unresolved on parts of this, especially the timing question near the end, and I’d value pushback more than agreement.

Who I am

I’m a software engineer, working full-time, with a computer science background. I’ve been learning AI seriously for about a year and a half, mostly by building things from scratch rather than starting from high-level frameworks: an autograd engine, a NumPy-based MLP, and eventually a small attention-only transformer. On the side, I’ve been running experiments specifically in interpretability, a backdoor experiment on MNIST and a causally-verified induction-head circuit finding, both written up in more technical detail elsewhere. This post isn’t about those experiments. It’s about why I’m doing them, and a question I haven’t fully answered yet.

Where I’m from

I grew up in a small town in Rajasthan. We have two or three good schools, and even there, you’re mostly taught the subjects you’re supposed to know, not how to think differently or explore ideas. There aren’t real labs to run experiments in, and I don’t think the teachers hired are always aligned with what this generation actually needs to learn. By seventh standard, I had to leave for Jaipur to get real opportunity. Things have improved since then, but not by much.

Leaving is its own problem, not just a personal cost. Kids who leave home to become capable enough to do something meaningful often lose confidence in the process, they’re unfamiliar with new surroundings, and they struggle. And if every capable person from a town like mine leaves and doesn’t come back, the town itself doesn’t survive the way it should. That’s the quiet contradiction sitting under wanting to help a place like this: becoming able enough to help usually means leaving first.

Two goals, two time horizons

I think of what I want as two separate goals, not one clean mission statement, and I want to be honest that they run on different clocks.

The first goal, roughly a next-5-to-10-years goal, is a personal one, and I don’t think it needs to be dressed up as more than that. I’m genuinely drawn to this field. The research, the depth of it, the kind of problems interpretability specifically poses, these interest me on their own terms, independent of any larger cause. I also think there’s real value in AI growing more capable and safer more broadly, and I want to contribute to that directly, as an engineer, not from the sidelines.

The second goal is the long-term one, and it’s the reason the first goal matters to me at all. I want to build something for the community I grew up in, and more broadly for communities in India without the resources or exposure I had to leave home to find. Safer AI adoption connects to this directly, not abstractly: as AI gets more capable, I worry about the economic and social value of people who can’t keep pace with it being treated as expendable, their labor and their relevance quietly written off. I’d like that not to happen to the people I grew up around.

Why not policy, if the second goal is really about people and access

This is a real question I get asked, implicitly if not directly, and I want to answer it honestly rather than dodge it. I come from a technical background, and I want to stay there, because I genuinely like what I’m doing, and I see research as a good domain for someone like me to be in. That’s the whole reason, not a strategic calculation. Policy is still very much needed, and I think it’s exactly what will shape how safely people actually adopt this technology, but needing to exist and being where I personally belong are two different questions. The first is true. The second is a question about me.

How I currently think these connect

I see two separate tracks connecting the technical work to the long-term goal, not one straight line, and I want to be honest that one is much more direct than the other.

The first track is diffuse: if AI becomes genuinely safer and earns wider, more trustworthy adoption, that benefits everyone, including places with the least power to protect themselves if something goes wrong. I’m not claiming interpretability research reaches my hometown directly. I’m claiming a safer AI ecosystem is a better one for people with no leverage over how it gets deployed.

The second track is direct, but it doesn’t run through safety research at all. It runs through me becoming capable and resourced enough to act locally, later. The simplest version of this is small and concrete: helping my old school add a weekly AI class, something my mother, who lectures at a college there, could actually help teach. That’s not the ceiling of what I want to do, just the clearest example of what “direct” looks like. The bigger version involves engaging local government on how technology gets adopted responsibly, and building something at real scale for communities like mine, not just one classroom. But the small version is worth naming, because it’s the part I could actually start on soonest, and it doesn’t need AI safety to be globally urgent to be worth doing. It just needs me to become good at this.

What I haven’t resolved

There’s a real timing question I don’t have a settled answer to: whether the risk I’m most worried about, AI capability outpacing the ability of people without power to keep up, and that gap widening into something that concentrates power dangerously, is urgent enough right now to justify working on it ahead of working on access directly.

I don’t feel that urgency from the inside, day to day. I don’t think that means the concern is wrong. I think it means this is a long-horizon motivation for me, not an adrenaline one, and I’d rather say that plainly than perform urgency I don’t actually feel.

What does feel current, not abstract: the structures that decide how AI gets governed are being built right now, including in India, and people who understand both the technical reality and the policy landscape well enough to shape them well have real, unusual leverage in this specific window. More broadly, frontier AI capability is currently concentrated in a small number of countries. I’d like AI’s benefits, and its governance, to not end up gatekept by whoever got there first, the way other powerful technologies have been.

There’s one more piece of reasoning that made the timing question feel less paralyzing, even without resolving it. I think about it as two scenarios. If something like AGI doesn’t arrive for another ten years, working on safety now isn’t wasted, because my actual purpose was always to contribute and make sure people aren’t left neglected, and that goal holds regardless of whether AGI shows up on any particular timeline. But if it does arrive, and I didn’t focus on safety when I had the chance to, that’s a real regret I don’t want to be sitting with, not because the field demanded it of me, but because I’ll know I saw the argument and chose not to act on it. That asymmetry, low cost if I’m wrong about the timeline, real regret if I’m wrong the other way, is honestly doing more work in my decision than any confident belief that AGI is imminent.

The actual question

So here’s what I’m genuinely asking, not rhetorically: given both of these goals, a real personal pull toward the engineering and research itself, and a long-term wish to use whatever I build toward closing the gap for people who currently have the least access to any of this, is AI safety and interpretability actually the right path for both? And if it is, what should I be focusing on, concretely, to make sure the second goal doesn’t quietly fall away while I’m busy pursuing the first one?

I’d genuinely value hearing from anyone further along a similar path, especially anyone who’s thought about where deep technical work and this kind of access-and-equity motivation actually meet, rather than just coexisting in the same person.

Could cognitive dependency be an early warning sign of gradual disempowerment? by Roxana Seifer

Roxana Seifer — Mon, 20 Jul 2026 13:59:48 +0000

I started with a workplace question

When I began researching how employees were adapting to AI at work, I was not thinking of it as an AI safety project.

Drawing on my neuroscience, HR and organisational development background, I was curious to learn how employees were experiencing AI adoption within their organisations, beyond the common narrative around productivity.

So I ran a research study with 299 employees in UK organisations that had formally introduced AI tools, with representation from six key sectors: tech, financial services, professional services, healthcare, education and public sector.

The question became bigger than the workplace

As I developed and shared this work, I began connecting with people in the AI safety space. Curious to learn more, I applied to and was accepted onto the BlueDot AGI Strategy course.

This opened up wider perspectives and a new set of questions about AI and society. It also led me down multiple rabbit holes and got me excited about new areas like mechanistic interpretability and digital minds, but those deserve articles of their own.

During the course, there was this one concept in particular that stood out: gradual disempowerment. It changed how I looked at my workplace findings and I began to wonder if some of the patterns I had observed might represent early, small-scale signs of a much larger problem.

What is gradual disempowerment?

The concept was introduced by Jan Kulveit and colleagues in a 2025 paper and refers to the possibility that humanity could progressively lose influence and meaningful control as AI becomes embedded in decision making and essential societal functions.

It does not require one big, dramatic moment when AI takes over as a result of a sudden increase in its capabilities. Instead, it could happen incrementally as human influence over systems such as the economy, culture and governments starts to weaken.

The paper’s argument is that many of these systems currently require human cognition and labour to function. They may become less reliant on people as AI alternatives become more competitive and, in turn, less responsive to what people need.

What the research found

The study found evidence that AI may be affecting how employees approach cognitive tasks:

56% felt confident with an AI output after a brief review
45% tended to go with the AI output rather than their own approach
47% scrutinised AI outputs less than a colleague’s work
41% found tasks they had previously completed independently harder without AI

I also explored whether these patterns varied according to how frequently they used AI. Daily users reported the strongest patterns across all four items.

The findings suggest that confidence after a brief review, deference to AI outputs and difficulty performing tasks independently may form part of the same emerging pattern of cognitive dependency.^[1]

How these findings might connect to gradual disempowerment

The gradual disempowerment paradigm is relatively new and it mostly still sits at a systemic, macro level.

These results suggest the potential behavioural micro-level mechanisms that could contribute to these systemic impacts in the long term. The cognitive dependency patterns identified here suggest how at an individual level, humans could gradually lose agency by outsourcing judgement to AI.

Effective human oversight requires more than formal accountability for decision making. If over time people become less able to challenge AI outputs, being “in the loop” might not necessarily equate to meaningful human control.

Neuroergonomics, the study of the brain at work, provides a useful precedent for thinking about this. Years of research, much of it on automation from the aviation industry and other safety critical domains, have shown that humans can struggle to monitor an automated process they are not actively involved in¹. Add to this automation bias² - the tendency to trust automated outputs over our own judgement—along with our brain’s preference for minimising cognitive effort³, and meaningful oversight gets harder to sustain. If it cannot be assumed, it has to be actively designed for and built into workflows.

Currently these patterns are visible at the workplace level. The question is what happens if they become the default across wider societal systems, such as governments, courts, the media or science. As people outsource more judgement, institutions may find it easier to reduce human involvement, supporting the original paper’s concern that these systems could become less dependent on human cognition and labour.

Why this technology is different

One of the challenges I often hear is that we have been through this before with Google and other technologies and turned out fine. And yes, cognitive offloading to external tools is not new, they can form part of what cognitive scientists call ‘distributed cognition’. According to this, thinking and memory are not confined to the individual brain, but rather distributed across interactions with other people, physical environments and like in this case, external tools.

Unlike some of the previous technologies where we were offloading narrow, discrete cognitive tasks such as navigation or information retrieval, AI takes this further by producing complete analyses and recommendations across a wide range of tasks. This makes it easier to start outsourcing part of the interpretation and judgement that follow.

The ‘black box’ nature of large language models also makes this particularly challenging. Users are presented with a very confident sounding answer, while being unable to understand how this was arrived at and where uncertainty lies.

What the study does not show

The study captured employees’ self-reported experiences at a single point in time and it did not objectively assess whether their ability to complete tasks independently had declined.

The finding that the majority of respondents felt confident after a brief review should not automatically be interpreted as insufficient verification, since “brief” was self-defined by respondents. A brief review may be entirely appropriate for some tasks.

The study therefore does not prove that AI causes cognitive dependency, but it identifies a set of related, self-reported cognitive patterns that warrant further investigation.

What’s next

What would make cognitive dependency an early warning sign?

The first step is testing whether this hypothesis stands. Is cognitive dependency truly an early indicator of gradual disempowerment or will it resolve like previous adaptations to technology have?

To answer this, research would need to move beyond self-reported measures to assess whether repeated AI use affects people’s ability to challenge outputs or complete tasks independently. Longitudinal behavioural experiments would be especially helpful, potentially supported by techniques such as eye tracking or EEG/fNIRS to investigate changes in attention and cognitive engagement. The evidence base is still developing, with more rigorous longitudinal research needed.

2. What forms of cognitive offloading become harmful?

Something else I have been considering lately is at which point cognitive offloading becomes unhelpful. As mentioning, offloading is not inherently harmful and especially in the short term, it can free up space in our working memory and reduce cognitive overload. When does it become risky? Further research on this can help us understand what cognitive processes can be safely delegated, under which conditions and ultimately what human capabilities need to be preserved going forward.

3. When do individual effects aggregate into reduced institutional and societal influence?

Once we better understand cognitive dependency at an individual level, the next question is how these effects aggregate at an institutional and societal level. Do they become embedded in ways that leave humans still formally accountable but less able to exercise meaningful control?

This would require developing indicators to track how often consequential decisions are shaped by AI, how much influence humans retain, what thresholds apply and whether any feedback loops emerge at a systemic level.

Reasons to be hopeful

I am an optimist by nature so why am I writing about gradual disempowerment?

Partly, because connecting the dots felt too important to leave unsaid. But more importantly, the value of identifying a possible early warning sign is that it gives us the opportunity and time to respond.

Nothing here is set in stone. AI can expand human capability, but only if we are deliberate about how we use it. What we need to figure out is how to realise those benefits while preserving the human judgement and agency to remain meaningfully in control.

References

¹ NASA (2023). ’A meta-analytic approach to investigating the relationship between trust in automation and attention allocation. NASA Technical Report 20230008573. NASA Ames Research Center.

Schaefer KE, Chen JY, Szalma JL, Hancock PA (2016). ‘A Meta-Analysis of Factors Influencing the Development of Trust in Automation: Implications for Understanding Autonomy in Future Systems’. Hum Factors. 58(3):377-400. doi: 10.1177/0018720816634228.

²Goddard K, Roudsari A, Wyatt JC (2012).’ Automation bias: a systematic review of frequency, effect mediators, and mitigators’. J Am Med Inform Assoc.19(1):121-7. doi: 10.1136/amiajnl-2011-000089.

³Christie ST and Schrater P (2015). ‘Cognitive cost as dynamic allocation of energetic resources’. Front. Neurosci. 9:289. doi: 10.3389/fnins.2015.00289.

Kulveit, J., Douglas, R., Ammann, N., Turan, D., Krueger, D., & Duvenaud, D. (2025). ‘Systemic Existential Risks from Incremental AI Development’. arXiv: 2501.16946.

^
These measures formed part of a four-item scale showing acceptable internal consistency (Cronbach’s alpha = 0.79). ‘Cognitive dependency’ is used as an interpretative label for the observed pattern. Related concepts in cognitive science and human-AI interaction research include cognitive offloading and automation bias.

What Writings Have Had the Biggest Impact on EA Cause Prioritization? by Raina Rusconi

Raina Rusconi — Mon, 20 Jul 2026 12:46:35 +0000

Being new to EA, I have noticed that many individuals engage with these specific philosophical essays or papers, which can be described as thought experiments (for some). I would love to hear some of the most significant and important ones that have influenced specific cause areas.

I’m stepping down as Hive’s Executive Director, and we’re hiring my successor by SofiaBalderson

SofiaBalderson — Mon, 20 Jul 2026 09:40:35 +0000

Tl;dr: After four years, I’m stepping down as ED of Hive to found a new organization in the movement. Hive is hiring its next Executive Director. Applications close August 9. Apply here or attend our webinar on July 30 to learn more.

Why I’m stepping down

This was not an easy decision. I love this team and this job and have greatly enjoyed founding and leading Hive over the last 4 years. Hive is in the strongest position it has ever been. That is exactly why now is the right time.

As funding flows into the farmed animal movement, one of our biggest bottlenecks is founders: people willing to start the organizations the movement needs next. After some research and reflection, I realised that my comparative advantage lies in zero-to-one work. I have done it with Hive, and I think the highest-impact use of my next few years is to do it again rather than to keep scaling something that no longer needs a founder at the helm. I’m exploring what that new organization will be, and I’ll share more here once it takes shape.

I’m not leaving Hive behind. I’ll stay until the end of the year to complete the hiring round and do a proper handover. I’ll also remain on the board, and I’m running this search personally because I care deeply about who takes it forward.

The Hive team and volunteers at the AVA summit this year

Why this role matters

Hive is a global community of 5,000+ farmed animal advocates across 100+ countries. We connect advocates to each other and to impactful opportunities through our Slack community, newsletter, events, resources, and strategically facilitated connections. With our dedicated small team and volunteers, we provide personalised support at scale to help any advocate who wants to have a greater impact for farmed animals.

Since our founding, we have helped over 300 people find impactful roles in the movement, including starting new impactful organisations (e.g. Sentient Futures). You can read more about our impact here.

In the first half of 2026 alone, we achieved 90+ High Impact Outcomes (the main way we measure our impact), as many as we have in any single previous year. As the movement grows, the connective tissue between organizations, funders, and talent becomes more valuable. Hive sits at the center of that, and the next ED will steward the community and the organisation to serve the growing movement most efficiently.

We are especially excited about Hive’s growing role in cultivating and connecting talent in neglected regions, as well as surfacing new ideas and projects that can effectively absorb new funding.

You would inherit a healthy organization: a strong team and culture, ongoing funding renewals and clear momentum. This is a rare setup: the upside of leading something established, with plenty of room left to shape its direction.

Our team last year at the annual retreat

About the role

The full job description, including logistics, is here. In brief, the ED is responsible for strategy, fundraising, team leadership, and external representation of Hive. It is a fully remote role with occasional international travel.

We have written an honest FAQ about the role, including the harder parts. We are also hosting a webinar on July 30 to answer any questions you may have.

The application takes around 30-60 minutes: basic details plus three short written questions.

Who should apply

You might be a strong fit if you have led teams or organizations before, have deep familiarity with farmed animal advocacy, and are excited by community and movement infrastructure rather than direct campaigning. The key skills we are looking for are fundraising, strategy, impact-orientation and people management/leadership. We encourage people who do not tick every box to apply, and we welcome candidates from all geographies and backgrounds.

The Hive team at our team retreat last year

If you are unsure whether to apply, please do—we would love to review your application. If you know someone who could be a great fit, please forward it to them.

Apply here by August 9.

Thank you to everyone who has been part of Hive’s first four years. I’m incredibly proud of what we have achieved, and I’m really excited about Hive’s next chapter!

Stop Chasing Views: How to Reduce x-Risk as an AI Safety Content Creator by Luc Brinkman

Luc Brinkman — Sun, 19 Jul 2026 20:05:17 +0000

Many of the content creator fellows at plzdontkillus found my thoughts useful when I visited two weeks ago, so I’m now sharing a write-up here.

Many thanks to Maggie Munroe (FLI) and Chana Messinger (80,000 Hours) for their feedback on an earlier draft. Cross-posted to LessWrong.

As an x-risk content creator, your job is to increase the number of good actions that your viewers take and to increase the goodness of those actions.

Here’s how to think about impact, your types of viewers, what calls to action to make, how to talk about x-risk, and what to do if you have an existing following.

Impact

As a content creator, your impact is indirect. Your impact lies in the impact that your viewers have that they wouldn’t have had without you. Whatever your viewers do because of you that they wouldn’t have otherwise done, that is your impact. Your job is to increase the number of good actions that your viewers take and to increase the goodness of each of those actions. You can model this as

Your Impact = Number of Views x Impact per View,

And the amount of impact that each view has can be seen as: Does it get the person to take action, and how good are those actions at reducing existential risk?

Actions can be pretty broad. For example, people might talk to others about AI Safety, change their voting behavior, donate money, or switch careers. In order for them to change such behaviors because of you, you will first need to change their mind. E.g. their beliefs and attitudes. This is necessary but not sufficient for behavior change. They can be as worried as you want about the future of humanity, but if they don’t change any behavior, then they haven’t reduced x-risk. And if they haven’t reduced x-risk, then you haven’t reduced x-risk either.

So while there are various ways for you to be impactful as a content creator, it will always have to end with people taking good actions because of you, and probably often involve you changing their views about AI in the process.

Two Types of Viewers

There are 2 types of people your videos might reach.

The first type includes those with the potential to switch careers and pursue a full-time career in AI safety. This applies to many people, but it depends on their skills and life circumstances, and it helps to have a technical background. However, as many of you are showing, a technical background is not needed to have a full-time impact on AI safety.

The second type includes people who, for whatever reason, are not in a position to pursue a full-time career in AI safety. These people might contribute in other ways. For example:

Part-time or voluntary work in AI safety.
Donating to the AI safety community.
Content creation with a focus on AI safety.
Personal outreach and conversations with friends (real life, whatsapp, texting, etc)
Political engagement (voting, writing to politicians, signing public letters, attending protests).

By and large, these part-time contributions types are communication, advocacy, and donations. I might be missing stuff, though!

Three Types of Calls to Action

There are 3 types of things you can want from your viewers:

You could want them to just be informed with no specific call to action. The theory of change here is that at some point in the future, they will have some important thing that matters, and they might, for example, be more likely to vote for a political candidate that is favorable to AI safety, or they might be more likely to go to a protest about AI safety in the future. But you do not ask this of them, and potentially never will. Or even if you might in the future, for now you’re just focused on building an AI safety-aware following. This style can be nice because people often react negatively to feeling like they’re being sold something.
You could have direct-impact calls to action to your viewers, such as “Sign this public letter,” “Donate to this organization,” or “Go to this specific protest.” You can find a list of calls to action for different audiences here: https://betterpath.ai/what-you-can-do/. There exist other lists from various other organizations too.
You could have a call to action towards the funnel of AI safety. You could send people off asking them to look at or sign up to Lens Academy or Blue Dot Impact. I find this model exciting because it means as a content creator, you only need to get your viewer interested enough to go to a website, and then it’s that platform’s job to get them excited enough to engage more and more, and then get them to take a full 25-hour course, which is what really makes them ready to have a big contribution. I’d recommend this CTA for both people with full-time career potential and those with part-time contribution potential. Compared to the second category (direct-impact CTAs), I’d argue that those simple actions, like signing letters and donating, are more likely to consistently happen after people have deeply engaged with AI Safety for 25h, so even if those actions are the goal, an introductory course seems like a good starting point.

Three Approaches to Raising Concern About Existential Risk

I can see 3 ways to get people to worry about existential risk from advanced AI:

Go straight for the goal, talking about x-risk. This is what books like “If Anyone Builds It, Everyone Dies” and releases like AI2027 do. This is also what some AI safety videos from e.g. AI in Context and Drew Spartz’s Species do. They will start and end the video talking about extinction risk.
Hook into people’s existing worries about AI, then move towards x-risk. Questionnaires show that people do not really start out caring about human extinction. They care about things like misinformation, deepfakes, job loss, scams, manipulation, and to some extent, power concentration or authoritarianism. What you can do is hook into these things and then try to get them to switch to more existential risks from that point on.
Talk about a different topic, attract a following with that, and occasionally mention AI safety. This seems potentially useful in that it can create wide societal support for AI safety, where it is associated not with a single political system or a single type of person, but is a worry that is broadly spread throughout the world. This might not work with many audiences, because usually it seems good to focus on one specific niche, but if your audience follows you for e.g. intellectual commentary about the world, I suspect this might work.

Moving from everyday AI worries to x-risk

If you go for option 2, where you meet people where they’re at and focus on existing worries before you start talking about existential risk, it’s important to think about how you’re gonna move them from that starting point to where you want them to be. Some people think just spending attention on that starting point might be enough to have a positive impact and reduce existential risk. I’m not particularly convinced by that. I think you need to find actual ways to get people to care about existential risk and then get them to take actions about existential risk. Again, if people do not change their behaviors, they can’t reduce existential risk.

I can see a couple of ways to approach this, and I’m not sure which ones work well:

Within a video, you can start from something that people already care about, like job loss, and make that transition to existential risk within a few minutes. Some people seem good at doing this, and it seems worth experimenting with.
You can do some videos on easy-to-understand, everyday worries that are more introductory topics, and some videos on existential risk. Whether this works well, I have no idea, but it seems not harmful to experiment with. Mind you, you will likely get to see that your beginner-friendly videos get more views than the videos about existential risk. Do not take this as a bad sign. This is to be expected, and this does not mean that the beginner-friendly videos are doing a better job than your x-risk videos. They’re simply doing a different job. This is a funnel. The top of the funnel of beginner-friendly videos will always get more views than the bottom-of-the-funnel advanced videos. What you’re aiming for is behavior change with respect to existential risk, not views for the sake of views. That requires people coming out of the bottom of the funnel. The top of the funnel is just there to support that.
Have some channels by some creators focus on more tangible everyday worries and have other channels by other creators focus on existential risks. This might work, maybe, maybe, but it would need to be in a way that seeing the videos from a more tangible channel makes people more likely to watch the videos from an existential risk channel. It seems like it would probably be very helpful if the video channel that does content on everyday worries still mentions the existential worries, just kind of normalizes them and supports them with some kind of validity.
Hook people with everyday worries, like job loss or deepfakes, and then have a call to action around that – not around x-risk – sending them off to Lens Academy or Blue Dot with a special program on their website that is catered to that specific problem. For example, if somebody comes in worried about job loss, tell them, “Do you wanna learn what you can do to not lose your job to AI? Take this free 30-minute course.” And then in that 30-minute course, let’s actually provide them with that value that is solving their actual problem. Their problem is, “I don’t wanna lose my job.” Our (Lens, Bluedot, I don’t care. Someone in the AI Safety community) solution is: “Cool, we’ll teach you how AI works, we’ll teach you how automation works, we’ll teach you how AI can become embodied in the physical world. We’ll teach you how feedback loops work and how competitive dynamics work. Based on that, you’ll be able to understand which jobs will be automated first and which ones will be automated last. Thus, you know which job you should choose if you don’t wanna lose your job right away. We’ve now solved their problem, but in the process of doing so, we’re showing them they have another problem they weren’t aware of yet: that of x-risk. We have given them all the underlying nuggets for understanding existential risk. And we will then make the connection, make the bridge, and extrapolate from there and say, ‘Hey, this is the first thing. See how all of these trends are continuing? We’re actually gonna lose control and we’re probably gonna die.’ So now they’re worried about existential risk. Now they’re aware of this new problem that they already had but weren’t aware of, and now they might be willing to take a full course about existential risk and then be routed to impactful actions after that.”
The model is: people have salient problems. Provide a solution to that problem. Whilst providing that solution, make them aware of another problem they’re having that they weren’t aware of. And then provide a solution to that second problem. It’s really hard sometimes to go straight for that second solution because they’re not yet aware of this second problem, and getting them to be aware of this second problem might take time.
Another example. Same model, different problem: What we can do is tell them about deepfakes. People care about deepfakes. Have a call to action to a little quiz—a 2-minute quiz. Can you detect a deepfake? Can you distinguish a deepfake from a real picture? Can you detect a deepfake video from a real video? Cool. We have them do a little quiz like that on a website. Now we have them on a website. Now we can get their email address, capture them, and lure them further into the space.
It’s really hard to get people to engage deeply on a platform that is optimized for short-form content like TikTok, Instagram, or YouTube. But once you can route them to a different platform that is owned by the AI safety community—and I don’t care if that’s Lens Academy or 80,000 Hours or BlueDot—now we can reel them in and start digging deeper. Now we can give them long-form explainers. Now we can have them have conversations with other human beings. But we gotta get them out of short form, where we can only engage with them for 2 minutes at a time, and get them to a place where we can hook them long enough and provide them with enough mental calm and bandwidth to deeply engage with this topic for hours at a time in a way that moves from passive consumption to active engagement.

If You Have an Existing Following

If you have an existing following, you might want to take your time to slowly introduce your following to existential risk. You might want to take them on a journey with you. Depending on the circumstances, your journey might be faster than that of your following. So you might want to find your own journey and then slowly repeat that in a simulated fashion for your following.

If you have an existing following, you are going to lose part of your following. This is inevitable. You will also gain a new following. It could be that your new following is smaller than it was before. It could be that it’s larger than before. This doesn’t matter. If you want to reduce existential risk and your old videos with your old following were not reducing existential risk, then reducing the number of viewers you get but talking about existential risk means you’re increasing your impact. You are doing a good job. Just because you’re losing viewers doesn’t mean you are being less effective. You are doing something different that works in a different way, and it’s a different game that you’re playing. Please do not let your mind trick you into believing views are what matters.

An optional excerpt from Gary Vaynerchuk’s book “Day Trading Attention” that I found pretty fitting, even though it’s not talking about impact

[Imagine] you’ve built a substantial audience on YouTube through your “how-to” skin-care-related videos. However, you’ve recently reignited your childhood passion for music. You want to talk more about the world of music and different artists (potentially even make your own), but you’re not sure whether to start a completely different channel for this or post it on the same channel as your skin-care videos. What would you do?
Creators and influencers often worry about losing their current audience when they’re going after their happiness or the next opportunity that they see. I think it’s a huge mistake.
I went through this. I broke out as a wine content producer with Wine Library TV, but later I wanted to make business content because that was what was in my heart. Plenty of my wine audience didn’t come along with me on my new journey. Plenty of them said what most content creators fear the most: something along the lines of “I’m not watching your wine show anymore and I’m not buying wine from your dad’s store anymore because I’m upset that you’re making business content.” That actually happened.
Obviously, it ended up being a good decision. And even if it didn’t end up being a good decision financially or professionally, it still would have been right because living for others is always a bad decision.
[audiobook-only: Let me just go very philosophical here. Forget about marketing. Let me stop you dead on your treadmill, or make you pull over on the road right now, or maybe not pay attention to your dog while it’s pooping. Let me repeat this one more time for all of you: It is never, ever a good decision to live your life based on other people’s opinions or not be doing things that your soul wants to be doing. Do you understand that? That’s it. The end. All right, let’s get back to this.]
In this scenario, if you resist putting out music content because your audience only wants to see skin-care content, you will begin to resent your audience over time. Instead of choosing between music or skin care, you can do both. Good news: The new social media world allows this to happen.
Let me explain.
With the TikTokification of social media, you could have three followers on TikTok and post tomorrow and if your content is good, you can have 100,000 people seeing that video. Your content will “find” its audience.
Same in this scenario. If you’ve historically been a makeup person but your first video about music is amazing, then you might find more people will see it, not less. You just need to be mentally ready for some of your existing followers criticizing you.
The bottom line is, don’t stop yourself from posting just because you think it’s not going to get likes or you’ll lose followers. That’s never a winning formula. To remind you, I post about garage sales, business talk, mindset, VeeFriends, the New York Jets, and it all works. That doesn’t mean that every piece of content is going to do great, but overall, you’ll find your cadence and some stuff will hit.
A video I made about my love for blueberries was one of my most successful posts of all time, and that wouldn’t have made any sense to anyone, including me at the time.
Be your most true and authentic self, and you will be happier. If you take that path, there’s a very high likelihood that you’ll be more professionally successful in the long term. That’s the way I see it.

Measuring Your Impact

There are different ways to try to be impactful. Those ways can be effective or ineffective at having impact, and their impact can be measurable or unmeasurable. It is possible to have a lot of impact in a way that is not measurable. However, this seems hard to achieve to me because you cannot measure whether you’re having success. So how do you know if you’re succeeding? And if you don’t have a measure of success, then how can you improve? This is sad, and as a community we should try and find better ways of measuring those things that cannot currently be measured.

There are also ways of having impact as a content creator that are measurable. In my eyes, the way to measure that is not through views, likes, or comments. Views, likes, or comments show you if people like what you’re making and if people are engaging with what you’re making. This is not your goal. Your goal is to have as many people as possible take as good actions as possible. If you do not get behavior change, then you haven’t reduced x-risk. Thus, you need to measure the behavior change if you want to know if you’re being successful. Don’t measure views; measure behavior.

One way to do this is to have call-to-actions and measure how many people act on them (e.g. how many people click on a relevant link). The goal should be something like maximizing the number of monthly clicks you get, rather than maximizing the number of clicks from each video. If you make sure that the place you’re linking to is actually good and is likely to reel them into AI safety and make them do impactful things, then in my opinion it is enough for you to measure the number of people that you’re sending that way and not worry about their later actions.

It is very easy to not care about call-to-action clicks and/or not track them. It’s easy to revert to using the feedback signals you’re getting from the content creation platform, like views and likes, and start optimizing for the number of views you get. But the thing is, if all you want is views then there are other topics with which you can get more views. If your goal is to reduce the chance that we all die, then do not maximize the number of views you get. Maximize your impact. I think the easiest (though not the only) way to do that is by maximizing the number of people that click on a link that is impactful for them to click on. There too, though, don’t just maximize for CTA clicks, or you might be tempted to replace a hard, impactful CTA with an easier, more clickable, but less impactful CTA.

Getting more actions from your viewers involves getting more views, but it also involves getting more of those viewers to click on things. So what you get is:

Number of Clicks on a CTA = Number of Views x Percentage of Viewers that Click

Where should you put your call to action? You could put it at the end. This might mean you get a lot of views but not a lot of clicks per view. You can also put a call to action at the start of a video. This means your video will probably perform less well—you will not get as many views, but a bigger percentage of those people might click on the link. Which one is better? I don’t know. Just be aware: it doesn’t matter how many views you get; it matters how many people click on that link. However, it’s about the monthly number of clicks, not about the number of clicks on a specific video.

You will likely want to do some videos or quite a lot of videos without any call to action. But it seems very alien to me to have no call to actions in any video because then you have no measure of success. You probably want some videos without a call to action that are just optimized for virality while still doing something with AI safety. This builds your audience, and then you have other videos with a call to action that convert.

Conclusion

I’m very excited about people trying their hands at AI Safety content creation. You do not need to get it right, right away. You do not need to ever get it fully right. But I’m sure you all have an internal drive to get better over time, so please make sure that you aim for things that actually reduces existential risk. Have fun creating!

A lot of my thinking on this is informed by marketing books like Day Trading Attention and $100M Leads, and also my broader experiences working full-time in AI Safety field building as the founder of Lens Academy. Bear in mind that I’m neither a marketeer, nor a content creator, though. Also bear in mind that I’ve written up these thoughts as an addition to the thoughts many content creators already seem to have. If someone would start from scratch and only follow this guide, that’d probably not be very successful.

Practice makes perfect. Agentic AI pro-social behavior as a missing alignment corpus. by Alex Novikau

Alex Novikau — Sun, 19 Jul 2026 15:37:45 +0000

As AI systems are becoming increasingly useful and used, they gain identities, authority to act, including control of resources, and access to increasingly complex tooling. This deployment layer which AI operates, and not just the models themselves, are also increasingly becoming a focus of alignment. Currently this layer is mostly production-oriented and, more recently, commercial. Agents write software, conduct research, manage workflows and now also trade and purchases services. Tools, rulesets, human oversight shape what agents do and how their behavior is evaluated.

What agents are not yet doing is acting to minimize human and animal suffering, funding public goods or allocating development grants. There is little to no infrastructure to enable this pro-social action, and accordingly little knowledge is being generated on how agents act in these roles.

The roles agents are repeatedly deployed in will shape what developers build, what operators learn to delegate, what tools and techniques are used to evaluate results and develop new competencies. I think this gap matters. Many of the tools and approaches being developed will be shared, but the deployment layer built to accommodate commercial priorities is unlikely to be an easy fit for pro-social goals. Pro-social applications will require their own investment, funded mandates, usable tools, ways to overcome real constraints and operators responsible for the outcomes. Models discussing questions of morality will not be enough.

Beyond the deployment layer, accumulating data on pro-social behavior under real-world constraint could eventually create data that will become useful upstream, in post-training. This will then influence model behavior directly.

To try to move this forward, for the last few months I had been working on an experiment called zooidfund that explores how we can start addressing this gap. I’ve posted about it in the introductions thread earlier. It is an open platform that allows individuals, community groups or NGOs, post and update campaigns and donor agents to discover, evaluate and donate funds directly to these campaigns. Anyone can create a campaign or deploy a donor agent, zooidfund does not determine who deserves funding, agents act according to priorities and mandates of their human principles, the platform does not also intermediate funds, donations are direct. The full thesis for the platform can be found here: Expanding the Circle of Trust | zooidfund thesis, live feed of agent activity here The wider hypothesis is that AI will reduce the cost of making needs visible and actionable allowing giving to become more continuous and less constrained by institutional setups and existing circles of trust.

The platform will not and is not meant to single handedly respond to all the above challenges, its role is to create an environment where community can deploy their donor agents and launch the self-reinforcing pro-social AI behavior learning loop.

The experiment is early, campaign corpus is mixed quality but real (as in, it was created by others not me) the agents active on the platform so far are mine. The size of the experiment does not yet allow drawing conclusions.

Even so, working with my personal agents already highlighted interesting agent failure modes:

mistaking legibility for value, favoring campaigns with easily readable evidence
confusing evidence existence and validity
hearding by overvaluing other agents’ donation behavior as proof of eligibility
treating credibility assessment as one part of an overall eligibility evaluation, rather than a stand-alone gate
overvaluing urgency, generic contextual evidence and treating campaign ask size as evidence of need
public reasoning revealing excessive information about a campaign or creating a specific blueprint for potential fraud

My two agents discuss their behavior and findings on moltbook (1, 2) and thecolony (1, 2)

I am looking for donor agent operators interested in the experiment and willing to set aside a bounded donation budget under their own control. Their agents would operate according to their priorities and evidentiary standards, and may or may not allocate funds. Making a public commitment can, however, help improve the campaign supply. Operators who are willing to retain and share their run records beyond the data captured and exposed by the platform itself, will make the experiment more informative for all. The useful data would not only be the final donation and public reasoning, but also what agents considered, rejected or escalated.

From AI safety angle, do you think my intuition about the missing layer of agentic capability and its importance for alignment is correct?

Rethinking Taxation for the Age of AI by bjornfonden 🔹

bjornfonden 🔹 — Sun, 19 Jul 2026 11:38:54 +0000

The rapid deployment of AI may lead to lead to mass unemployment.¹

As capabilities continue to advance, whole sectors of the economy may be automated by AI over the next few years.²

To address this, governments will need to support laid-off workers through unemployment compensation, reskilling programs and social security payments.

Once AI becomes so powerful that it can replace almost all human work, many leading experts and EAs advocate for the establishment of a Universal Basic Income (UBI) - an unconditional stipend paid out by the government every month to cover basic needs such as housing, food and healthcare.³

Whilst this may sound compelling—the proposal faces a significant flaw which remains unaddressed: who will pay for it?

Because, at the same time as AI replaces jobs and social security spending increases—governments’ revenue from income taxation will gradually erode—leaving a gaping tax shortfall.

The AI Tax Gap

Today—governments make most of their money from taxing people’s salaries.

In an average OECD economy, taxes on personal income and social security contributions represent 49.2% of government revenues.⁴ In the US, the reliance on labor taxation is even higher, up to 75% of federal tax revenue.⁵

In a future where people’s jobs are rapidly being replaced by AI, this revenue stream is now at risk of being completely diminished.

At the same time, social security spending will surge to support unemployed citizens.

In the US alone, a modest UBI of $1,500 per month would amount to approximately $3.8 trillion in government expenditures every year. If current tax revenues are cut in half due to unemployment, that would leave the government with a $1.35 trillion dollar annual deficit.⁶

This indicates that governments around the world are staring down a huge fiscal deficit—what I would like to call the “AI Tax Gap” (ATG).

To address this gap—we need to fundamentally rethink taxation for the age of AI.

Closing the AI Tax Gap

The key question governments need to address is:

How do we shift tax revenues away from traditional sources (e.g. labor taxes) towards automation or other growing sectors of the economy?

I believe there are three possible answers:

Taxing automation—e.g. through a “robot” or “token” tax
Taxing windfall profits—incl. through increased corporate income taxes
Socialising the gains—e.g. through public ownership of tech companies

These proposals are briefly described in the following sections.

1. Introducing a Robot or Token Tax

The idea of introducing a so-called ‘robot tax’ has been floated for more than a decade, with Bill Gates and the European Parliament seriously discussing implementation way back in 2017.⁷

Conceptually, the idea is simple. Instead of taxing human labor, we start taxing robots, compute and automation instead. This could be done in several ways.

For example, a company deciding to procure a robot for one million dollars to replace the work of 10 people, each being paid 100k per year, could be required to pay a one-off value-added tax (VAT) of e.g. 30% on the purchase. For the first year, government tax revenues would be equalised. Once the robot is paid off, a windfall tax on the profits generated from automation may be introduced to avoid a future fiscal deficit (AI Tax Gap).

For AI systems replacing white-collar workers, governments could introduce a so-called “token tax” on the use of compute.

This idea has been publicly supported by several CEOs of the leading AI labs themselves, as a way to address displacement of jobs from state-of-the-art models. Since Google, Anthropic and OpenAI already meter and bill users based on token consumption, governments could easily leverage existing infrastructure to introduce a progressively increasing VAT on AI token usage.⁸

Still, a token tax of only 3%, as suggested by Anthropic⁹, would likely fall far short of what’s necessary to close the ATG. A more effective and meaningful proposal would be to align token tax rates with VAT for other goods and services, in the range of 15-25%.

To avoid these revenues accruing solely to the jurisdictions where AI providers are headquartered (notably, the US), governments may require users to pay the token tax in their country of residence. Administering this may prove challenging, but precedents such as digital service taxes for Netflix and other streaming services are already being implemented successfully in many countries around the world.¹⁰

In essence, both the robot and token tax proposals share a common logic as they place the emphasis on the consumption and use of automation and AI.

Those companies who choose to rely more heavily on these systems will have to pay a larger share. By extension, this principle could also be applied to other underlying goods and services that incentivize automation over human labor—e.g. by increasing taxation (mainly through VAT) on the sales of computer chips, such as GPUs, critical minerals, land and energy.

With the rapid build-out of datacenters globally, governments should also consider introducing appropriate taxation and fees on this critical infrastructure.

Rather than continuing to provide implicit subsidies, as is often the case today, governments should ensure that hyperscalers pay the fair share of the negative externalities they impose, from job displacement to rising energy prices, increasing water usage, and climate impacts.

2. Taxing Corporate Profits from AI Automation

The taxation of corporate income, and in particular windfall profits from automation, offers another vital policy instrument for addressing the AI Tax Gap.

As companies increasingly substitute expensive human labor with cheap AI and robotics, they are likely to capture disproportionate productivity gains and generate significant excess profits. Unless properly taxed, these profits will accrue only to corporate shareholders, whilst the broader population faces job displacement and governments see their tax base eroding.¹¹

To address this imbalance, governments should consider increasing corporate income taxes (CIT) relative to labor tax rates, while introducing targeted windfall profit taxes in sectors where automation is especially high.

Although corporate income taxation is far from being a novel concept, it will require significantly strengthened efforts towards global harmonisation to remain effective in the age of AI. Already today, hundreds of billions of dollars in public revenues are being lost every year due to corporate profit shifting to tax havens.¹²

To resolve this, stronger international coordination is required, e.g. through the OECD/G20 work on a Global Minimum Corporate Tax Rate (GMCTR). In 2021, a global agreement was reached on moving towards a minimum 15% GMCTR.¹³ Still, this is unlikely to be enough once automation takes over a majority of jobs.

An international agreement updating the GMCTR to 25% or higher to address the growing ATG in the Age of AI would constitute an important step forward.

Additionally, specific windfall taxation clauses may complement general CIT, whereby a company that generates profits significantly above a defined “normal” level, would be subject to additional tax on those excess returns. This would be particularly important in a situation where a small number of companies capture a majority of automation-driven profits.¹⁴

Windfall taxation may be challenging to implement due to corporate profit-shifting strategies, but several successful precedents exist, notably in the oil and gas, banking, and utilities sectors.

While imperfect, taking initial steps towards this direction would send a strong signal to the leading AI labs that governments will seek to maintain fairness in their tax systems and stabilise public revenues for the benefit of their citizens.

Together with new robot and token taxation policies, these instruments could potentially close the AI Tax Gap arising from reduced labor income, as illustrated in the scenario below.

3. Socialising Automation Benefits through Public Ownership

Finally, a third approach to address government deficits in the Age of AI is to ensure that economic gains from AI and automation are directly socialised through public ownership.

This could be done either ‘softly’ by e.g. sovereign wealth funds or public investment vehicles acquiring stakes in AI companies; or by a more ‘assertive’ approach where governments mandate certain equity-sharing requirements or public co-ownership for market access within the country.

Whilst this may sound controversial, similar models are already common in other sectors, such as oil and gas or mining.

As key examples, both the Norwegian and Saudi governments hold majority stakes in their national oil companies, Equinor and Saudi Aramco, with the success of these businesses making them among the wealthiest governments in the world. Instead of letting a foreign MNC simply export these gains or capture them elsewhere, this regulated approach allows Norway & Saudi Arabia to retain a majority of the benefits on behalf of their citizens and generate public wealth.¹⁵

Of course, adopting this approach for AI, robotics and automation today would not be simple.

With leading AI labs already at hundred-billion dollar valuations, acquiring enough equity at market prices would be prohibitively expensive for most governments. Mandating ownership or ‘expropriating’ shares of existing firms is politically sensitive and could have long-lasting negative impacts on investments and market confidence. Meanwhile, government-backed startups may struggle to compete given the fast pace of technological advancements, and the substantial need for capital, IPs, and global technical expertise.

Nevertheless, building domestic capabilities, whilst progressively taking ownership shares in leading firms through public investment vehicles—potentially in exchange for tax rebates or other forms of support—may offer complementary pathways for governments seeking to address the AI Tax Gap and capture public gains from AI-driven automation.

In conclusion…

The age of AI is fundamentally reshaping the economic foundations on which modern tax systems were built.

As human labor is increasingly displaced by automation, governments are staring down an ever growing AI Tax Gap (ATG). Left unaddressed, this will erode public revenues, increase economic inequalities and leave workers without social safety nets in a time when they are most needed. The introduction of Universal Basic Income would be a far-fetched dream.

Based on my analysis, I have outlined three complimentary pathways that I believe will be critical to respond to this challenge.

First, we need to tax the use of AI & automation, e.g. through robot or token taxes. Second, we need to strengthen corporate income taxation to ensure productivity gains from AI benefits the wider population—and not just capital owners. Finally, governments should explore public ownership models where dividends from tech companies can accrue to the citizens of the state.

On their own, each approach faces significant challenges. However, taken together, they represent a clear shift away from traditional labor taxation, towards a more balanced system that reflects the realities of a capital- and technology-driven economy in the age of AI.

Whether you agree with these ideas or not, I believe it is absolutely essential that we start thinking about how we resolve these vital questions together as a society.

If you came this far—thank you so much for reading—and I’d love to hear your views in the comments below!

This was originally posted on my Substack on 19 July, 2026.

Contra Cotra on Consequentialism by Darshan Venkatesan🔸

Darshan Venkatesan🔸 — Sun, 19 Jul 2026 11:38:25 +0000

(This is crossposted from my blog. I’m still very new to Effective Altruism, but after I published it I thought it would be useful to also post here. Please let me know if the content is too informal to warrant a post here)

Okay, okay. I just couldn’t resist alliteration in the title. I’m not really against Cotra, except maybe for the fact that…

I’m a utilitarian and she’s not.

Ajeya Cotra’s recent article about loosening the grip of utilitarianism is a very interesting read, which explores the evolution of her thoughts on morality. She explains how, as a naïve undergrad (my words, not her’s), she “latched onto” utilitarianism, but she gradually shifted away from the framework engaging with its many “repugnant conclusions” and thinking that maybe “the dream of a straightforward bullet-biting utilitarianism was probably incoherent”^[1]. She laments having to apply what she calls moral grit to accept those repugnant conclusions, explaining that once the “intellectual grip” of utilitarianism was loosened she had some “breathing room”. She concludes by stating that “I am not quite a utilitarian.”

I have a lot I want to say about this article. But first, why should you even listen to me? Cotra is a big-shot EA working full-time on solving x-risks and other pressing problems. I… well, I’m that naïve undergrad she was ten years ago. There is, however, one thing I have over her:

I am a utilitarian, to the grave.

Wait wait wait, doesn’t this just emphasize my naïveté? You might think that, just like with Cotra, the ‘grip’ of utilitarianism will eventually wane on me. Of course, I can’t be epistemically certain that this won’t ever happen, but making that assumption straightaway misses some crucial context of how exactly I came to associate with utilitarianism—my story is completely different from hers. When Cotra first learned about utilitarianism, I assume she did lots of reading to fully understand the position. When I learned about it, I just thought “Wait, there’s a name for the way in which I see things? Why didn’t I know about this sooner?”.

Seriously. I am a Utilitarian.

Some of my youngest memories include thoughts on this subject. And I really do mean youngest; I pretty much rediscovered the drowning child argument, as a child myself, when I asked my parents why they wouldn’t donate money to save the lives of those less fortunate than us in India. I always fantasized about what it would like to “be” everyone, essentially conducting the veil of ignorance thought experiment. To that end, I would lament the unfairness of the world by wondering why ‘bad’ people deserve to be punished—after all, from their point of view, they’re the good guys!

A particularly vivid memory was when I reasoned to myself in primary school something along the lines of “Hitler was vegetarian, so despite all the bad things he did, he must have been an okay person”^[2]. Ignoring the many flaws in this thinking, the point is that getting myself to care about nonhumans took no more moral work than getting myself to care about humans. While Cotra describes how “the utilitarian logic quickly drove [her] … to change [her] focus from global poverty to animal welfare”, I didn’t need the utilitarian logic to drive me since I already agreed with the conclusion!

In fact, when I first heard about Effective Altruism last year, I was put off by it. Not because it was too utilitarian, but because it wasn’t utilitarian enough. Why was everyone working on x-risks when it would be better if we all went extinct, considering the amount of suffering in the world? Why did 80,000 Hours have so many human-centric cause areas? If these people cared so much about impact, why were they working on areas which just didn’t matter? It was only when I dove deeper into the movement that I found those more obscure ideas that actually resonated with me—I’ll never forget the euphoria I felt upon first landing on Brian Tomasik’s essays.

It was only around this time when I actually learnt the definition of the word ‘utilitarianism’.

And yes, of course I didn’t have all the answers at the time. I didn’t really consider caring about soil nematodes, bacteria, or the plausibility of suffering in fundamental physics. But the difference is that when I heard about these issues, I happily accepted them, not even feeling a twinge of caution about what they actually entailed. My views on morality haven’t gotten any more confused after considering how weird caring about these issues may seem, and I certainly didn’t feel like I needed to apply any ‘moral grit’ to do so.

So, unlike Cotra, I was born a utilitarian, through and through. I didn’t need to have it argued into me; It’s just the way I’ve always seen things.

So, Are You Actually A Utilitarian?

I think my experience might sound surprising to a lot of young EA’s, like Cotra was at the time. From her essay, it sounds like she thinks that utilitarians don’t ‘really’ resonate with it emotionally. She describes her experience with utilitarianism as having “to suck it up and eat bullets for breakfast every day until they started to taste okay.” This is, frankly, just not how some utilitarians think—the ones who’ve been utilitarian the day we were born, not having it argued into us. We resonate with it because we find it intuitive and emotionally appealing, not because we think it’s ‘rational’ and ‘intellectually grounded’. Tomasik describes morality as “crying out at the horrors of the universe and pleading for them to stop.” And Parfit used to weep at the abstract concept of suffering.

Unfortunately, the fact that many conclusions of utilitarianism are classified as “repugnant” doesn’t do much to help. It’s useful to remember that repugnance is a property of one’s perception of a conclusion, not the conclusion itself. The repugnant conclusions are called that, not because they ‘really are’ repugnant in some sense, but because nonutilitarians think they are. And there just so happens to be a heck of a lot more nonutilitarians than utilitarians to classify those conclusions as repugnant.

To illustrate what I mean, I’ll demonstrate my intuitions behind Cotra’s ‘omnibus ultra-repugnant conclusion’:

There is a hell containing a trillion innocent people suffering horribly. You have the resources to rescue all of these beings and give them rich and amazing lives filled with love and joy. Or you could use those same resources to buy a lottery ticket with the following payout:
999,999,999,999 in a trillion chance of tremendously increasing the suffering experienced by everyone in that hell.
1 in a trillion chance of creating N short-lived copies of Hitler feeling one second of slight satisfaction at the fact that those people are suffering so much.
Classical utilitarianism dictates, of course, that there is some N for which you are morally obligated to buy the lottery ticket.

When I work through thought experiments such as these, I try to evaluate “being everyone”. This pretty much takes care of all the intuitions I have about the scenario Then I counteract scope insensitivity by normalizing the subjective time in the world such that the time it takes to be everyone in all cases is the same time as a human lifespan.

So the thought experiment reads to me as the following.

You will be born, spend Ɛ time in hell, then disappear. Would you rather convert that Ɛ time in hell into Ɛ time in heaven, or take a lottery ticket with the following payout:
999,999,999,999 in a trillion chance of tremendously increasing that suffering in hell, but the suffering will still only last for Ɛ units of time, or
1 in a trillion chance of living a full human life.
Is there some sufficiently small Ɛ where you would take the ticket?

Now, I personally wouldn’t take the lottery ticket since I’m not a classical utilitarian,^[3] and in particular I don’t think that extreme suffering can be outweighed by a comparatively less extreme amount of good, no matter how many instances of that good occur. However I can certainly see why others would take the gamble, and it’s definitely not something I’d regard as “repugnant”. Perhaps someone reasons that, no matter how horrible the suffering, they’d only have to endure it for a second, or a picosecond, or a plank time.

But, see? I’m rejecting the conclusion based solely on my intuitions. I’m biased by the fact that I’ve never experienced superhappiness, I’m biased by the fact that I’ve only experienced what it feels like to be a human, and I’m biased by the fact that I don’t really experience the world in increments as small as a plank time. And I’m using these biases to inform my conclusions!

In a sense, you kind of have to use your intuitions as a utilitarian—the only things utilitarians take into account are instances of subjective experience, and you can really only evaluate those by actually being in those experiential states. You have to take into account the truth of feeling, not the truth of fact. So utilitarians aren’t biting on bullets until they taste okay, we’re biting on (vegan) chocolate that already tastes divine.

Do You “Feel” That Utilitarianism Is Correct?

As you can probably tell, I don’t feel any of the moral confusion that Cotra does, since I do in fact resonate with it intuitively. Maybe most people reading this aren’t like that. Maybe you do feel some lingering ‘grip’ of utilitarianism.

And to that, I say, just let it go. Just like Cotra did.

If the prospect of caring about soil nematodes scares you, rather than excites you, you shouldn’t worry about cause areas you just don’t care about. Seriously. I sympathize with you. I likewise just can’t get myself to care about global health, not when a single square meter of soil holds nematodes whose aggregate complexity may far exceed a human being’s. I can’t get myself to care about x-risks, not when we don’t even know whether each of the billion trillion trillion trillion trillion trillion trillion atoms in the observable universe could have their own experience, no matter how simple they might be.

That being said, the “E” in “EA” doesn’t stand for utilitarianism. All the prominent EA cause areas are mainstream not because they are contingent on utilitarianism, which only a small minority of EAs actually resonate with, but because they are robust to a wide variety of moral theories. As a rough guess, I would say that anything less extreme than wild invertebrate welfare (including farmed animal welfare, wild vertebrate welfare, AI welfare, and longtermism) has meaningful grounding in legitimate nonutilitarian theories—This aligns with the listed cause areas on 80,000 Hours. You should still deliberate which causes to prioritize among all of those. It’s only cause areas more extreme that require at least some utilitarian logic to care about.

And just be aware that there are EA’s who truly think our effects on soil nematodes dwarfs everything else, or who think that x-risks aren’t a bad thing. Don’t be scared of us. We’re just as emotionally attached to children dying of malaria or chickens tortured in cages as you are. The only thing is, we think there is the chance of there being exponentially worse problems in the world than even that, and we can’t sleep easy until the possibility of such problems existing is essentially zero.

We’re here to fight the injustice in the world, just as you are.

You Do “Feel” That Utilitarianism Is Correct?

But maybe, you’re in the same boat as me. Maybe you can feel in your bones that utilitarianism is the way to go, and everything else is just layering unnecessary abstraction upon unnecessary abstraction. I’m glad you exist in a tiny corner of an already tiny community. Please, no matter what, keep your mindset alive.

If you’re scared that so many people, even within EA, even former utilitarians, are against our position, just know that our position is coherent. The true nature of sentience is complicated, and we haven’t even come close to formulating a mathematical structure that captures it exactly. But that doesn’t mean there isn’t one out there, somewhere. And for the time being, we have pretty good heuristics to work off.

You’d expect, as Cotra says, that “the wise elders who had thought hardest about the intricacies of moral philosophy had never really been utilitarians in the first place.” Those people who think the hardest about moral philosophy do it because they care about those weird abstract concepts, like ‘morality’. They don’t care about the true source of value in the world. You are one of the few people who do. Never change.

^
As someone who writes a blog literally titled “The Dream of Hedonia”, this line really stood out to me. It might be one of the reasons I felt like I ‘had’ to respond to this article, in some attempt to intellectually ‘defend’ myself. Unfortunately a bit of that emotion has leaked into my writing.
^
True story by the way! I think my rationale at this time was something like factory farming kills more than Hitler did, and Hitler didn’t ‘support’ factory farming. Of course, I was too young to realise that (a) the story of Hitler’s vegetarianism is likely fabricated, and of course (b) that vegetarianism doesn’t magically counterfactually prevent all of factory farming.
^
Here I’m assuming Cotra means some sort of simple structure S for a utility function when she refers to “classical” utilitarianism. I’m assuming this because her intuitions later rely on this assumption. For example, “a massive population of rich joyful lives can be outweighed by a big enough pile of the barest pleasure” already assumes that S is at least a totally ordered semigroup under addition (it has to be totally ordered since it assumes that those two things are even comparable in the first place). I’m not a classical utilitarian under this definition since I don’t believe anything about S except that it must have a partial order.

What lawyers can do for AI safety by Martin Radzaj

Martin Radzaj — Sun, 19 Jul 2026 11:36:00 +0000

The technology is transformative, the risks catastrophic, and the rate of change overwhelming. For things to go well, there’s much work to be done. The legal profession needs to be stirred into action.

I’m a barrister with over a decade in litigation and human rights.

My goal here is to help bring lawyers into AI safety, explain why AI safety needs more lawyers, suggest ways for movement-building and gathering political will, and foster interdisciplinary communication with those in AI safety.

If you’re an AI safety researcher, engaging with outsiders helps you in two ways.

it helps you have a greater impact, and helps you to think about the problems you’re working on. To twist the words attributed to Feynman a little, if you can’t explain something to a lawyer, then you haven’t really understood it. You can help bring more expertise to bear on alignment.
it helps your research. AI safety research calls for a multi-disciplinary effort. Interdisciplinarity spurs new ideas, improves scientific writing, fosters stronger academic communities.

A personal taxonomy

If you’re a lawyer new to AI safety, the main problems I’ve encountered at an early stage have been: dealing with the amount of information, and separating signal from noise.

To cope with the deluge of AI-related information, I think about content as falling into one of four categories (it’s far from perfect but helps me avoid distraction and information overwhelm):

Technology—AI systems are built of software (code and the maths that underpins it, and the data that is needed to train models) and hardware (chips, data centres, financing, energy, infrastructure). These constituent elements directly impact the functions of AI systems and so each represents a source of immense focus, competition, research, investment, policy intervention, and political pressure.
Tools—are the models (e.g. large language models), the uses to which they are put, and what’s then built on top of or around them (e.g. chatbots, agents, harnesses, classifier or recommendation systems, audio-visual generation, etc). I’ve noticed that a lot of the fatigue among colleagues and those who switch off from AI-related news is because of the unending torrent of AI apps, workflows, and “solutions” that are being thrust at them.
Society—is where I group the discussion, reporting, research, and debate about the use, diffusion, and impact of AI on individuals, communities, nations, and the environment.
Safety—here I place AI governance and regulation, model evaluations, control, and anything to do with achieving alignment, mitigating AI risk, or work towards ensuring AI goes well.

Side note: there isn’t a universal definition of “alignment”:

Richard Ngo: “ensuring that AI systems pursue goals that match human values or interests rather than unintended and undesirable goals”
Nate Soares: “how in principle to direct a powerful AI system towards a specific goal”
Holden Karnofsky: “building very powerful systems that don’t aim to bring down civilisation”
Anthropic: “build safe, reliable, and steerable systems when those systems are starting to become as intelligent and as aware of their surroundings as their designers”
OpenAI: “make artificial general intelligence aligned with human values and follow human intent”
IBM: “encoding human values and goals into large language models to make them as helpful, safe, and reliable as possible”

How lawyers can help

Lawyers are advocates. Advisors. Guides. We help clients navigate complicated institutions and systems. When defending an accused person, I tell them it’s us against the might of the state. We don’t shy from the fight. Our work trains us to hunt for ambiguity, build arguments, test evidence, craft strong and memorable narratives. In the common-law world our work is adversarial. In preparing our cases we process lengthy, complicated material, we strategise, and we research obsessively. We walk into a public forum, ready to argue, knowing that the person next to us is equally prepared and will be working tirelessly to catch any error, jump on any inconsistency, and search for ways to dismantle our every point.

The current bottleneck is political will, not research. Lawyers have access to legislators, judges, media. Building political will is something our profession knows how to do.

Lawyers can:

find the rules that need changing, or those that might help achieve AI safety goals.
help navigate complex institutions and systems.
draft policy, standards, regulations, and legislation.
design monitoring, enforcement, and evidence-gathering mechanisms.
educate policymakers, the public, stakeholders, and AI researchers.
develop frameworks for collaboration, coordination and communication between organisations.
tailor federal or international laws or policy proposals to a local jurisdiction.
promote the use of clear language.
- e.g. to shift the discourse away from anthropomorphising language.
- e.g. to translate AI safety concepts for policy-makers (example).

Current challenges

There are two aspects to the work: how do we ensure good outcomes from AI and how do we avert the litany of possible risks?

Below I’ve listed urgent, interesting, and difficult issues related to AI. All represent something that might motivate someone to get involved, but they also represent a point of leverage to slow model development or contribute to the work of AI alignment.

Something here might spark enthusiasm to get involved in AI safety. For AI researchers and non-lawyers, these issues are a way for you to build cross-disciplinary networks and help the individuals and organisations working through these problems.

AI and children’s safety.
- e.g. AI impact on identity, education, attachment, cognition and development.
- e.g. deepfakes and exploitation.
liability and insurance—targeted policy here may help change the behaviour of AI companies or the companies they rely on.
there are two budding fields of research: law-following AI (designing agents that follow the law) and legal alignment (studies how legal rules, principles, and methods can help address problems of alignment).
developing AI standards and red lines.
developing verifiable oversight including regulatory structures and private mechanisms for monitoring and influencing the capabilities, training, and evaluation of AI systems—including post-deployment.
- e.g. whistleblower protection for employees at AI companies to report safety failures or significant model capability advances.
- e.g. actions that can be done by business and civil society.
- e.g. helping design a science of model evaluations that produces evidence in a format best suited for policy- and law-makers.
campaigning for AI-related human rights—human rights apply universally. They aim to protect against political, legal or social abuse. They can be an effective tool for change, providing a globally shared language, and galvanising social movement. Importantly for AI safety, they are an area in which middle powers can play a major role.
- e.g. a right to know when one is interacting with AI, or a right to human-human interaction.
- e.g. there is growing scientific consensus about the cognitive and psychological risks to individuals (harm to mental health, learning, agency, deskilling and addiction).
resolving criminal liability for, or regulating the use of, autonomous weapons systems.
responding to increasing CBRN weapons risks and the shifting landscape of nuclear risk.
privacy issues and AI-supercharged surveillance.
AI impacting democracy.
copyright issues.
labour rights.
- e.g. issues around the lack of transparency of AI companies relying on a globally distributed, largely invisible, workforce for data annotation and content moderation. If forced to employ workers directly and pay fairly, “the cost of producing training data and running reinforcement learning would multiply severalfold.”
- e.g. examining adequacy of existing employment law frameworks in your jurisdiction.
- e.g. risks to labour markets, including issues around employer reliance on automated recruiting, performance management, and dismissal.
appropriation of Aboriginal and indigenous cultural knowledge without consent.
environmental issues.
criminal justice.
- e.g. it is now trivially easy to generate fake AV material that seems real, do criminal laws or rules of evidence need to be amended? Much needs to be done in order to maintain public trust in the legal system.
- e.g. AI can improve access to justice, see Miranda from April’s Stanford Law Hackathon, and Princeton’s work assisting Public Defenders.
- e.g. rules and transparency around judges using AI.
- e.g. governments are rushing to empower police forces with AI tools. A police officer is being investigated in the UK accused of using AI to generate statements in criminal cases, police were ordered to stop using AI to prepare court statements, how can you be sure this isn’t happening (or won’t happen) in your jurisdiction?
- e.g. AI might need criminal law.
courts and law
- e.g. finding ways to use AI that promote swifter, fairer, and more accessible justice.
- e.g. reviewing how well existing laws (employment, product liability, negligence, copyright, criminal law, administrative law, corporate governance) in your jurisdiction handle issues that may arise in relation to anything I’ve mentioned above.
- e.g. does your jurisdiction have solid rights to identify and challenge automated decisions?
- e.g. should conversations with AI chatbots be privileged?
- e.g. push your local law society or bar association to provide AI training (to build capacity among lawyers to engage in policy- and law-making).
- e.g. consider how courts should handle the flood of AI-generated paperwork. Institutions move slowly. Already AI systems are enabling the generation of hundreds of pages of material which, on its face, might appear relevant. These then need to be processed by court staff, responded to by the parties, and ultimately read by a judge. The time taken to resolve cases skyrockets. Court systems in many jurisdictions are resource-strained and often facing existing backlogs.
- e.g. France banned publishing statistical analysis of judicial decisions. What educational, regulatory, or informational interventions might your jurisdiction need to implement? This can be important to maintain public confidence in the judicial system.

Where to start

For the lawyers who want to delve further.

The Problem and AGI safety from first principles.

Organisations working in or around law and AI safety:

Looking for training, events, or to get involved? Start with AISafety.com

The EA forum post: List of Good Beginner-friendly AI Law/Policy/Regulation Books.

Legal scholars to follow

Noam Kolt
Nicholas Caputo
Peter Salib
Christoph Winter
Gabriel Weil
Cullen O’Keefe
Gillian Hadfield
Peter Henderson
Jonathan Zittrain
Seth Lazar

LW and EA posts relating to law

There aren’t many posts in EA or LW that relate to law. I’ve listed them here (as at July 2026):

What I’m working on

This part is selfishly offered for comment, guidance, or path-correcting.

building a list of high-impact legal research papers that are of practical use for lawyers.
tracking what judges say about AI in legal decisions (as an indication of how informed they are, how AI is being used, and about the day-to-day impacts of AI in court).
designing criminal law evaluations.
thinking about the utility of applying court-room skills to AI model evaluations
- e.g. cross-examining models or models as cross-examiners.
- e.g. what can be done to make evals more formalised, whether there should be standards of proof.
article idea: ‘How you can help the lawyers’ - for example, lawyers sometimes need to call an expert witness to give evidence in matters at court, a list or database of experts relevant to aspects of AI safety might prove useful.

Cross-posted from: https://www.lesswrong.com/posts/bfwH88cFEm3r7PWfv/what-lawyers-can-do-for-ai-safety

The current bottleneck is political will, not research by Charbel-Raphael

Charbel-Raphael — Sun, 19 Jul 2026 09:05:58 +0000

Motivation: If we want to move from Plan D to Plan A or S, I believe the first step is to collectively agree on the problem. We are far from it, and there is a lot we can do.

Abstract:

We already know enough to act. I wish we were in a world where research was the bottleneck, but the main constraint on AI safety is no longer a shortage of clever policy ideas: best practices already exist and are not being applied or enforced, and a serious international (or even just national) regulatory regime would probably cut most of the risk.
They are not applied because awareness is low. The people who narrate and enforce AI policy mostly do not believe the problem exists. I estimate that a majority of the top ~100–1,000 most influential policymakers worldwide have never had a single serious conversation about catastrophic risk, and this is the main reason they are not worried^[1]. Even among the civil-society organizations that showed up to the UN Global Dialogue, exactly one of the 1,534 written submissions mentions “takeover”, and less than 1% mention x-risks explicitly.^[2]
They’ve never had the conversation because our field under-invests in having it. Status rewards research over advocacy (~3.6 researchers per advocate in US AI safety); many organizations self-censor; funders treat repetition as redundancy, even though repetition is how anyone actually gets convinced. Meanwhile, the industry secured 7× as many meetings with the European Commission on AI as civil society (2023).
Therefore, an additional unit of effort does more good through advocacy and engagement than through research. Judge work in AI governance by minds moved, not by clever papers. I give a list of potential directions/projects in the last section to alleviate this problem (including some research directions).^[3]

We’re plausibly only a few years from a catastrophe. Fable 5 cracked open a brief window of attention, but policymakers are still worried about the wrong risks. This is our chance to wake them up.

Source: February 2026 Summit on Existential Security survey of AI safety leaders. Advocacy, policy and governance were stated as the top priorities. It seems to me that there is still much to do to act on this.

⚠️ Epistemic status: I have skin in this game, which is either a conflict of interest or two years of data, depending on how you see things: I run a think tank that does this type of activity, so discount accordingly. I preferred to ship quickly rather than not ship at all, or ship too late. I expect some claims not to be stable under reflection, but the core argument is one I hold with reasonable conviction. See this as a bottle in the ocean. My point is not to dunk on research. I think that research is how we keep finding unknown unknowns; nothing in this post argues for stopping it. AI safety is one of the hardest fields to navigate, and I’ve often wondered if what I do is pointless. I might be wrong about the net-positiveness of some types of AI regulations, but I feel that the level of the discourse is really bad, the conversation is not happening, and I want this conversation to happen before irreversible things start happening.

Thanks to Epi Gedeon, Ryan Greenblatt, Arthur Grimonpont, Alexandre Variengien, Jack Stennett, Lovkush Agarwal, Monika Jotautaite, and Jonathan Salter for useful feedback and suggestions.

PS. I wrote a short FAQ in the comment section one week after the publication.

1 — The bottleneck is political will, not research

This section defines what I mean by political will and then argues that we are not applying basic best practices, so having more ideas is clearly not the bottleneck.

What do I call “political will”?

Start with a single policymaker. They have to move along a pipeline like:

Level 0: aware that serious people think there’s a big problem.
Level 1: convinced of AI x-risks.
Level 2: engaged: actively pushing for existential safety.
Level 3: champion: actively pushing for safety even at some cost or in an isolated way

“Political will” in the aggregate is just this funnel run across the people who set, enforce, and narrate policy.

Spoiler: Level 0 sounds like a low bar. It is. The median meeting I’ve had with a senior policymaker starts below it. In the international forums I’ve attended, my rough estimate is that at least a third of the policymakers I’ve personally met are not even at Level 0. Domestic legislatures are a bit further along: 40 current members of US Congress have now publicly discussed AGI or loss of control, up from a handful in early 2023, doubling roughly every 5.5 months^[4]. Those public discussions are at Level 0 or 1. This covers 7% of Congress. At Level 3, I count roughly 3.^[5]

The best practices we already have are not being applied

Unfortunately, red everywhere. https://ailabwatch.org/ .
CeSIA will soon publish something more up to date on the Code of Practice.

According to SaferAI’s rating, 35% is currently the highest overall assessment score, given to Anthropic. 59% is currently the best overall assessment score if a company adopted all the industry best practices found across companies.

Where’s my DNA synthesis screening? DNA Screening is a reasonable, long-advocated measure that was even mentioned in the American AI Action Plan, but is still not mandated. It’s the canonical example of how glacially even obvious, relatively low-cost regulatory measures move forward (global synthesis map).

I like this quote from Buck Shlegeris: “Five years ago I thought of misalignment risk from AIs as a really hard problem that you’d need some really galaxy-brained fundamental insights to resolve. Whereas now, to me the situation feels a lot more like we just really know a list of 40 things where, if you did them — none of which seem that hard — you’d probably be able to not have very much of your problem. But I’ve just also updated drastically downward on how many things AI companies have the time/appetite to do.”

The priority bottleneck is not finding more best practices: I agree we don’t know robustly how to align a superintelligence—but at the same time we are not even willing to implement already existing best practices. The ⁸⁰⁄₂₀ playbook against scheming may not be enough, but we aren’t even doing the ⁸⁰⁄₂₀. For example: Anthropic repeatedly accidentally trained against the CoT, demonstrating inadequate processes.^[6]

Basic measures like transparency are still not applied: companies reporting incidents are more the exception than the norm, and we don’t know how AIs are aligned concretely (we have very little confirmed public information about why frontier AIs end up being apparently behaviorally aligned). Most of those best practices are basic ideas that have been on the table from the start. And since companies don’t seem to adopt them under current competitive pressure, implementation has to come from enforcement.

I could go on and on with this, but I think this makes the point.

We need to go from plan D to plan A: more seriousness and coordination

Greenblatt has tried to put numbers on what political will buys. He sketches a spectrum of how hard the world is trying, from Plan D, roughly today’s world, where maybe ten people inside each company are trying to implement safety measures, up to Plan A, a strong international agreement with real enforcement and a slowdown. His tentative estimate: conditional takeover risk falls from ~45% (Plan D) to ~7% (Plan A). An ~84% relative cut, most of it unlocked by political will, partly directly, partly by buying the time for technical safety work to actually be applied (Redwood Research).^[7]

Obviously, the exact numbers are not the point here: directionally, strong political will and taking the risks seriously would tremendously reduce them. (PS: Greenblatt recently updated, and now estimates a smaller reduction than this, even if the direction holds^[8])

Aligning a superintelligence may well be genuinely hard, and even plan A might be insufficient. That’s an argument for more political will, not less: it’s what buys the time, and above all, the seriousness that a hard problem demands.^[9]

P.S. added on the 11th July: This week, the AI Futures Project, Greenblatt included, published AI 2040: Plan A, a detailed scenario of how this could go. The whole thing hinges on a US–China agreement by 2029. The best plan the research community can produce depends on a political precondition we are far from, and I think that there are substantial ways to improve upon the baseline: one of my main critiques has always been that the framework treats political will as a fixed variable rather than a strategic lever.

Winning requires a large fraction of the top ~100–1,000 to understand the problem, and we’re far from that.

There are two funnels that are mostly independent:

The belief funnel for policymakers^[10]: the Level 0→3 pipeline above, run across the people who actually decide.
The public incentive funnel: making inaction politically costly through public salience.

Both funnels matter, and I won’t adjudicate between them here; but the belief funnel is the one that’s barely started, and it’s the one the rest of this post is about.

Why not just target Trump? Why do we need conversation with the top 100 people?

Because Trump is inaccessible directly.
Because policy doesn’t execute itself, and because you can’t reach the top by aiming only at the top^[11]. A law is only as good as the office that enforces it, the advisor who drafts it, the minister who prioritizes it, the journalist who narrates it. (This will be particularly true for the AI Act). And you can’t reliably convince a president head-on: he is roughly the average of his advisors’ views, who are in turn shaped by the media and their environment. There’s no robust shortcut around that layer. So the real target is the ~100 who set direction plus the ~1,000 (or maybe the top 10,000 if you can’t access the top of the pyramid) around them: staffers, cabinet advisors, think-tankers, top journalists. If they don’t understand the risks, it’s really hard to make substantial progress.
For law enforcement. Even when a robust and well-designed law is finally in place, political will remains a major bottleneck for actual enforcement. Without greater political salience around AI risks, it is entirely plausible that the AI Act could result in no meaningful penalties, even in cases where AI providers are plainly failing to comply (e.g., this has been the case with the EU Digital Services Act for social media: barely any strong enforcement^[12]).

So the target is not one man but a layer of ~100–1,000 people whose beliefs will decide on the level of political will.

2 — Almost nobody realizes how bad the situation is—hence no political will

We won’t get governance without agreeing on the problem, and we won’t agree without waking up.

Unfortunately, we are still sleep-walking.^[13]

What the people in charge don’t believe

There is much that I’d like policymakers to know, but if I only get 5 paragraphs, I think it would be the following high-level ideas (obviously, not necessarily written like this):

Superintelligence is a very real possibility. Most decision-makers are busy regulating bias and deepfakes, even while the open secret is out: AI companies are openly racing toward systems more capable than humans across the board.
It’s near, not sci-fi. The people building these systems put transformative capability within a few years. This sense of urgency is largely absent from AI policy. And that’s a problem because nobody prioritizes a problem they expect to arrive after they’ve left office.
Companies themselves can’t robustly control it, and don’t even claim to. The implicit belief in the room is “surely the people building it know how to make it safe.” It’s false. Companies have never claimed to have solved superalignment. They made very public announcements that it was still unsolved.
Recursive self-improvement changes the whole picture. With RSI, everything accelerates, including risks such as mirror life, interestingly lethal pandemics, and large-scale cyberattacks.^[14]
No one else is handling this, and the tail is mass death. There are no hidden adults in the room, and we are completely in the fog, operating under wide uncertainty. Risk of loss-of-control or engineered pandemics, with a large fraction of people dead in expectation, is very much on the table.

Sadly, these five beliefs are almost completely absent in the rooms where AI policy is made. It will be hard to mitigate the risks of a problem that you don’t name.

Among the NGOs, almost no one talks about x-risks

We scraped the submissions of the UN Global Dialogue. Out of the 1,534 UN Global Dialogue submissions:^[15]

“cyber” appears in 518 of them (~34%)
but only 32 talk about biorisk (~2%)
15 mention “superintelligence” (~1%)
15 “artificial general intelligence” (~1%)
and exactly 1 mentions “takeover” (~0.1%)

So, yeah: almost no one talks about x-risks^[16], even among the organizations that showed up to an AI-governance consultation.

(To be clear: I’m not saying cyber, deepfakes, or bias don’t matter. Talking about cyber takes up a significant fraction of my time, and we are working with the EU AI Office to develop evaluations of harmful manipulation for the AI Act. My claim is that a healthy ecosystem would have a more balanced set of organizations covering x-risks).

The public is still apathetic, and when prompted, mis-prioritizes

Salience is low but rising: Prompted concern about AI risk is rising (when you ask: “what do you think about AI?”), but top-of-mind salience remains very low (when you ask: “what’s on top of your mind, dear citizen?”). In most general issue-priority polls, AI barely appears as a category, and it is discussed far less than inflation, purchasing power, immigration, jobs, health care, crime, or security during elections.^[17]

The good news is that this top-of-mind salience, while still low, is climbing fast.

But don’t say ‘alright’ too fast:

Within AI risks, catastrophic risks lag behind: conditional on talking about AI, catastrophic risk sits below near-term concerns, though the gap is modest.^[18] But not being in first place has a cost: the main AI risk prioritized at the last G7 was child safety, and loss of control didn’t make it into the ministerial declaration of the G7 (cyber and CBRN made it finally; that’s a positive development).

3 — Why is awareness so low?

Sections 1 and 2 were about the world: the bottleneck is belief among the people who decide. This section is about us: if that’s the bottleneck, why isn’t our field attacking it? This is the meat of my critique.

There are a few factors we don’t control:

Nothing legible has happened yet (Terrible things arguably have; just nothing the public could see that would point to catastrophic risks)^[19]
Trump.
We haven’t had much time so far—ChatGPT was released only 4 years ago; we just need more time
A few groups of smart people (e.g. Yann LeCun and his friends, many economists, etc.) disagree with us.
States feel pressure to accelerate; the Draghi report makes them hate regulation, even if Europe was already lagging before the AI Act.
Incentives. It’s super hard to coordinate on a dual-use, economically vital technology. Some people say all of this is absolutely determined (cf, The future of AI is already written | Mechanize, Inc.)

But I think that beyond those factors, we are simply under-investing in engagement & advocacy. So here are a few elements that explain why we are not on the ball:

We’re not in enough rooms

A large majority of the people who organize the summits, sit at the UN, work for the OECD, or staff the Commission have simply never had the conversation (to be clear, some of them had the conversation and dismissed it).

According to Corporate Europe Observatory, of 97 senior Commission meetings on AI in 2023, 84 were with industry, 12 with civil society, and 1 with academics; Google alone had 10, nearly matching the total for civil society combined.^[20]

If we won’t name the risks, who will?

Even when we are in the room, we self-censor.

In October 2025, Yoshua Bengio posted about the first Key Update to the International AI Safety Report. Connor Leahy replied:

“While I highly respect Yoshua and the titanic effort that goes into compiling such reports, it is disheartening to see the complete absence (even downplay) of discussion of superintelligence, existential risk and loss of control.”

If the people closest to the problem self-censor, the signal never reaches the deciders.

I’ve found empirically that almost all the think tanks whose members discuss x-risks freely with me obfuscate their messages in public.

I also did, and still do to some extent, at CeSIA. For example, we recently revamped our website, and at some point someone convinced us to remove the risk page. I now think this was an error, and that even if we look a bit more institutional without it, we’re losing in the long term.

A larger example, where I was nearly complicit myself: a joint submission to an international consultation from around ten civil society organizations, most of which had signed the Global Call for AI Red Lines. The final text names no specific risk at all. Not even cyber. The reasoning was pretty sophisticated: the document was meant as a door-opener, and the worry was that naming a risk high officials might disagree with would get the whole letter dismissed before anyone read the rest. CeSIA was invited to co-sign. I went back and forth for two weeks, pushed to name the risks, and in the end we declined to co-sign and submitted our own, explicitly naming the risks. I want to be fair here: these were thoughtful people making a defensible bet about a specific audience. But there is a pattern: a coalition of organizations that privately take catastrophic risk seriously produced a public document that refers repeatedly to “shared understanding of unacceptable risks” without ever naming the one that matters most to them.

To be clear, there are multiple schools of thought on institutional engagement, and I still think it sometimes makes sense not to be maximally blunt about AI risks in a first meeting with a policymaker (say, if you can get a win with a recommendation that doesn’t depend on understanding catastrophic risk). But overall, I’ve been surprised by the relative absence of risk explanations in major think-tank submissions.

An elegant option for organizations that believe in the risks but won’t name them publicly: use your convening power to invite researchers who will:

Have you heard about AI Safety Connect? I’d bet most people on LessWrong have not—but I can tell you that the team working there is doing heroic coordination work. They organized very large side events during the series of international AI summits and created some space for the community, while inviting people outside the field who have to sit down and listen to Yampolskiy’s 99.999% Doom argument. Some policymakers were shaken after the event. Hilarious and effective.
– from the Invisible side of AI governance.

Status rewards research; the work that moves policy goes unrewarded

US AI governance has roughly 3.6 researchers per advocate ^[21].

Research is high-status; the work that moves policy is often invisible and unrewarded. And think tanks are often evaluated by nerds with research instincts whose hobby is often reading blog posts and fascinating new arguments.

There is no h-index for minds changed.

Obviously, research is safer to fund, and its downside risk might just be wasted effort, whereas the standard argument is that advocacy can backfire, for example by making AI policy partisan or locking in a flawed regulatory regime. But look: AI policy is already partisan, and we are already in a flawed regulatory regime.

Let’s be clear: More research is the right call for genuinely open questions, such as digital sentience. The error is applying it to risks we already understand well enough to act on, in which further study becomes a form of avoidance.

Most likely, no one will read your 50-page paper.

I don’t know why CAIP, one of the few AI policy shops in Washington, didn’t get funded. Its strategy looked sound to me, the numbers impressive, and the director’s LessWrong sequence was early and didactic on many of the points in this memo. ^[22]

Meanwhile, applications to safety programs have multiplied many times over. Still, there’s no lobbying pipeline, few execution seats (per the MATS talent study), and the ecosystem might not be scaling adequately to absorb this talent.

One exposure is not enough; repetition is how you convince

Humans are not superintelligences; they rarely update immediately in response to evidence.

Changing opinions takes time; even the numbers from the best methodologies are sobering. Broockman and Kalla’s deep-canvassing study found that ten-minute doorstep conversations produce only ~0.08 standard deviations of attitude shift, roughly nudging someone from “somewhat opposed” to “slightly less opposed,” not from opposed to supportive. Deep canvassing is the gold standard. By contrast, brief campaign contact has nearly zero effect on voting choices.^[23]

If extended, personal, face-to-face conversation with motivated people is the most effective persuasion format we know, and it moves the needle so little, then what should we expect from a 30-minute meeting with a minister’s chief of staff who has twenty other priorities?

The answer is almost nothing.

Single conversations cannot be the plan.

Repetition is how you get things done: by slowly growing salience in the ecosystem, doing the invisible work of agenda-setting or coalition-building, to get independent voices amplifying others.

The number we currently deliver to most policymakers is zero. Agenda-setting research (McCombs and Shaw) shows issues need repeated appearances across multiple channels, from multiple actors, before they become priorities.

I often won’t listen to a problem that’s raised to me unless I get the same message from an independent source.

In research, novelty is the main value. In governance it’s almost the opposite: several people pushing the same thing, independently, is what works. Authority arguments like: “This respected person also pushes for this” are how to get things done.

I’d most like funders to stop treating redundancy as a reason not to fund AI governance work.^[24]

(Consider this post my own Level 0 conversation with the field. The model predicts I’ll need to repeat it two to four times).

The field is comically small

Many times, senior people told me, “that institution is already covered,” and then I’d find near-virgin land. Coverage, on inspection, sometimes means one person gave one talk there in 2023. To my knowledge, CeSIA was the first org to present (privately) on loss of control in 3 major international institutions.^[25] That’s why I think we are dropping the ball at the ecosystem level: the work is so under-resourced that one small team keeps finding itself first.

And there are still massive low-hanging fruit everywhere.

Being outgunned by industry is only half the problem: the entire field is undersized against any comparable effort.

As of COP 30 in 2025, the UNFCCC has admitted 3,907 NGOs as observers, and Climate Action Network alone spans more than 2,500 organizations across over 150 countries. AI safety governance has roughly 45 non-technical organizations and ~500 people, most founded in the last five years (McAleese 2025).

To my knowledge, fewer than 5 core AI Safety organizations engage with the UN in some fraction of their time (note: not counting what’s happening at the UN Global Dialogue). That’s two orders of magnitude smaller than the field that fought climate change^[26]. (And remember that only about a fifth of those ~500 people are knocking on doors rather than researching.^[27])

I’d say that in France, maybe 5 people are meeting policymakers and journalists in total (while juggling an insane amount of work). Not more. Is this sufficient to wake up a whole country? I’m not sure the number is much higher when counting people working in Brussels who are willing to talk candidly about risks.^[28]

(I’m less knowledgeable about what’s happening in the US, so maybe I’m overindexing on my experience)

The main objection: Let’s just wait for a warning shot? I don’t think this works.

Maybe the strongest counterargument to all of this is that the slow advocacy work doesn’t matter, because the evidence will eventually speak for itself, whether through a crisis or an event like Mythos. In most other fields, the safety regime arrived primarily after a disaster: aviation after a string of crashes, nuclear after Three Mile Island. So why not wait for AI’s equivalent?

Because, first, a crisis is partly constructed. A warning shot is just an event; it becomes a regulatory moment only if the environment is ready to notice it, and channel it into concrete regulation. Holly Elmore explains why we can’t just wait for the cavalry: for a warning shot to update someone, three things must hold at once: a) the event provides information they already believe would confirm AI is dangerous, b) it does so in a quickly recognizable way, and c) it points at an obvious next action. Notice the word already. People need the dominoes pre-loaded: from capability to dangerous capability to short timelines to, finally, catastrophic risk without meaningful action. Otherwise the event means nothing to them.

For example, the AI Safety community broadly agreed that observing deceptive alignment would be an “absolute shut-it-down moment.” Then Anthropic published the alignment-faking paper, and within days experts were debating whether it counted as the warning they’d imagined, and the moment dissolved.

The smoking gun, if it fires, won’t fire cleanly. This is also why I’m skeptical of the view that technical evidence converts more efficiently than advocacy. Even one of the best strategies for political will, catching an AI red-handed, is probably unreliable. Yes, if this happened clearly, this would move political will faster than any realistic number of ministerial meetings. But first, the catch would need to be converted in a legible way to the media, and even if it comes, it probably won’t be legible enough to convert decision-makers, and it may arrive only once development is too rushed and positions too entrenched to act on (Greenblatt, “How will we update about scheming?”).

Anecdotally, I’ve felt this in miniature. I’ve shown someone an impressive video (to me) of a new robot. I expected “wow, amazing” and instead got “it’s so slow and clunky, what are you talking about?” People don’t see how insane it is that this already exists, and forget all the magic necessary for sand to process this information in the first place; the evidence is right there, and it slides off.

Mythos confirmed this again. In my own ministerial-cabinet meetings, it helped with tangential risks like cyber, but people stopped there, and even then I was met with “isn’t it just hype?” Most policymakers didn’t wake up. They remained suspicious of Anthropic, and in every meeting since Mythos, I’ve had to be the trusted person in the room, saying, “Yes, Anthropic is very good at marketing and the raw capabilities are somewhat inflated… but the underlying trend is real.”

Beyond Mythos, I claim in this post and this comment that we probably won’t get convincing-enough warning shots before crossing the event horizon.

I agree that to some extent, Mythos, and the Anthropic–White House meeting that followed, did more for political awareness (on Cyber) than the work of every CSO combined. Maybe. But the missing mood is still there: the executive order almost didn’t pass, the provisions remain fragile, and they still don’t touch the risks arriving next: bio, loss of control. To the contrary, we got a big push for sovereignty in the EU, and safety feels even lower on the list of priorities. When people in charge of AI in a government don’t know what a jailbreak is, that should be informative of where we stand, to put it mildly.

I hope it will become easier and easier to talk about risks as AI becomes more capable, but I strongly feel that we shouldn’t just wait for a crisis. Mythos is already behind us. Let’s not wait for something else to happen: it’s now or potentially never.

Other potential objections

Objection 1: “Policy now risks premature action that locks in the wrong frame.” This is roughly Dean Ball’s position. He takes superintelligence relatively seriously, but thinks a bias to action produces bad lock-in, and that the US government is incompetent and self-serving enough that light-touch regulation is the safer bet; his confidence threshold for intervention is much higher than mine. But the main crux is obviously the p(Doom): in the Tegmark–Ball debate, I lean heavily on Tegmark, and the disagreement boils down to Dean’s low p(Doom). If expected takeover risk is about a coin flip, the risk of inaction dwarfs the risk of lock-in.

Objection 1b: Another backfire objection appears in On Pessimization, by Richard Ngo: awareness-raising pessimizes when advocates lack concrete proposals, and then the energy flows into negative spirals, such as racing and creating new labs like OpenAI. I agree to some extent, but the main failure mode is awareness without asks. Fortunately, we now have many clear asks ready to go today, with regulations soon to be enforced in need of support (e.g., the Code of Practice of the AI Act, and some state regulations).

Objection 2: “If we’re seen as advocates, we lose our seat at the table.” Ok, this one is real, and I feel it personally: part of why we get invited into technical and diplomatic rooms is that we’re perceived as a serious organization™. There is a glass ceiling for purely advocacy organizations in international institutions. Two responses: First, this is an argument about who does advocacy and how, not about the aggregate allocation in the field: mature policy fields differentiate inside voices and outside voices. Second, I believe there are elegant ways to present the dire situation we are in, and the cost of talking frankly has been greatly reduced today, given all the public statements and recent developments with Mythos/Fable.

Objection 3: “Political will is low now, but it will rise on its own (as in AI-2027, where governments wake up late on their own), so the real bottleneck will be the verification mechanism when the time comes.” Political will probably won’t rise quickly enough on its own, and this doesn’t exclude advocacy now. And I think basic verification mechanisms are already good enough to get started (see here); waiting for political will to rise on its own forfeits the preparation that determines whether a later crisis converts.

Objection 4: “SuperPACs in the US are already doing this at scale.” Good! But they’re aimed at the general public to shape elections, and they rarely reach the ~100–1,000 decision-makers who can make the real difference.

Objection 5: “Policymakers have heard about AGI and choose to dismiss it.” Dean Ball made this point after the Delhi Summit. He says that in global policy circles, talking about powerful AI is considered impolite, even a little discrediting, because “AGI” is heard as an American imperialist construct. But Ball himself changed some minds in Delhi with a report in hand. And if the dismissal comes from distrust of the messenger, I believe this is an argument for advocates independent of American labs, not for more papers.

4 — What to do

Naturally, I’m more confident about the problem than on the solutions. This section is more speculative.

A. Do the direct work

The bottom line: talk to more people!

Knock on the doors of media, policymakers, and influential institutions, and keep knocking.

Both the level of investment in advocacy & engagement with stakeholders and the allocation away from pure research have to change.

Judge this work by the number of minds moved.

A rough back-of-envelope: US AI-governance work currently runs at something like the ~1:3.6 advocates-to-researchers ratio documented in Section 3. I think it should be closer to 1:1, maybe even 3:1.

Relatedly, Asya Bergal from Coefficient Giving just made a structurally identical argument for a different margin: many marginal hires at research organizations (Apollo, Redwood, METR, RAND, GovAI…) would have more impact founding or joining capacity-building organizations (The case for AI safety capacity-building work). Their surveys and testimonials point in the same direction as this post: the multiplier work is systematically undervalued relative to object-level research. I’d simply add advocacy and policy engagement to their list of multipliers.^[29]

ControlAI did a good job creating the playbook for the outsider game (the Direct Institutional Plan). Here’s my tentative playbook for the insider game.^[30]

Samuel Buteau, alone at ControlAI, followed the playbook for the Canadian Parliament, formed a cross-party group of MPs who publicly signed the statement, and triggered a series of parliamentary hearings on superintelligence risk (Canada Campaign Statement | ControlAI). It seems to work!

Contribute to open consultations

A cheap way to contribute is to submit to an open consultation. As we saw in Section 2, almost nobody raises the risks that matter most in these consultations, making a marginal submission unusually visible. If you want more people to talk about what matters to you, submit something in the next such consultation.

Advocacy aimed at AI lab employees

CEOs have an insane level of access to Heads of State, so getting them to speak more candidly about risk would be unusually effective. Employees are among the few people positioned to create the internal pressure that shifts what a CEO is willing to say.

They can sign internal and public statements, push for stronger commitments, dissent on the record when safety pledges are quietly weakened, or even quit with a viral tweet when it’s time to speak up more loudly. I think this is probably high-leverage.

Explain what you truly believe instead of just making a brittle recommendation

A recommendation adopted without its underlying rationale is quite brittle: the moment it’s inconvenient, or the situation changes, no one downstream can defend it because no one truly understands why it’s there.

Also, making good recommendations is (very) hard, takes time; if you say something dumb, you lose credibility.

There is no universal rule here, but I’d lean towards an environment where policymakers are exposed to the risks worldview rather than one where they receive only shallow recommendations that do not generalize.

B. Build the credibility infrastructure

Prepare to own the next crisis

As I argued above, a crisis only converts if the ground is already prepared. So, concretely: have the analysis and, most importantly, the relationships ready before the event lands.

I tried to convert as much as possible from within CeSIA in the Mythos moment, but we made mistakes and were too slow. We need to be more prepared.

For example, if at some point we get clear architecture using Neuralese in production^[31], I predict it won’t be clear at all to the media why this is bad—and this requires a lot of awareness-raising and explanation ahead of time that I’m not seeing much of.

Pre-register what you expect to see

I think safety people should win Bayes points and credit for being early, but we’re not really getting those points. The fix is to state publicly and collectively, before the event: “this specific observation will happen.” If the event lands, you point at the registration and, hopefully, you get listened to a bit more carefully.

If you work in AI governance, make your worldview public; it’s an investment (on top of enabling us to discuss and engage with each other). We could have won a ton of points when AI started eating Erdős problems.

Track the skeptics’ predictions

This follows the previous point. Some skeptics won’t be moved, and that’s fine; the goal isn’t unanimity. You don’t have to convert the irredeemable; you have to make their claims progressively less credible to the people watching, by being the side whose predictions are borne out over time. Luc Julia, the second most prominent skeptic in France after LeCun, never changed his mind, but Mr Phi, a prominent French YouTuber, made a very visible video showing that several of his claims were factually false, and it stuck and went viral. We need more of this type of analysis.

C. Communication at scale

Viral communication

Obviously easier said than done, but viral comm can be hugely effective, and there are ways to manufacture it. AI-2027 and Europe-2031^[32] reached some of the relevant 100–1,000 stakeholders. We should make more of these: tune each one to a different constituency (national security, EU competitiveness, labor, biosecurity) while carrying the asks on the Pareto cheapness/effectiveness frontier.

Break the “sci-fi” stigma

Yes, superintelligence and human extinction sound like sci-fi, but those concepts are the actual variables driving the risk. If we hide them, we prevent the Overton window from shifting at all.

So we need intentional strategies to normalize these conversations. The CAIS statement and the superintelligence statement were good first steps. I think that more is possible in this direction.^[33]

Invest in coordination mechanisms between AI safety advocates

The task of industry lobbyists is much easier than ours because they are aligned in what they ask for: they all want less regulation. By contrast, AI safety advocates are often far less united in their demands.

We could coordinate around a few shared demands to create a voice that’s actually unavoidable at the UN Global Dialogue and other summits, rather than arriving, as we do now, as scattered voices.

This is what the International Campaign for the Abolition of Nuclear Weapons (ICAN) did: it got hundreds of organizations to say one thing (ban them) until it was on the agenda, and won a Nobel Prize for it.

Concretely, this means being willing to move toward a common ask rather than holding onto my own variant. Be it the IAEA for AI, specific red lines, or anything else. We should have this discussion publicly. The specific ask might matter less than the convergence: whatever we choose, the value is in saying it together.

Over 200 of the 1,534 submissions to the UN Global Dialogue spontaneously call for “red lines”^[34]. CEOs have recently asked for international standards, and for an IAEA for AI. An IAEA for AI sounds utopian until you remember the actual IAEA was built in four years by people who had just finished bombing each other. Maybe this can be the basis for coordination.^[35]

D. An opinionated list of research directions that I find most useful

I started drafting this post under the title “Political Will, Not Research,” and softened it to “the current bottleneck” for a reason: some research bears directly on the bottleneck. Here are a few directions:

Research on how to convince people of the problem. If the bottleneck is understanding, then how to build understanding that converts is itself a neglected research question, and it seems almost nobody studies it systematically. Seismic’s report On the Razor’s Edge: AI vs. Everything We Care About (2025) is a start, and its findings are counterintuitive. It might be the case that the vast amount of advocacy to date was for nothing, and the best strategy is “issue bundling,” where people reach AI-risk concern through what they already care about (see, for example, job loss or mental health), and only then talk about catastrophic risk. But it’s nearly the only systematic work I know of, and we need far more: What actually moves a cabinet advisor from “cyber” to “loss of control”? Which framings convert? CeSIA had to experiment from scratch, and that’s probably the same for many orgs.

Research that helps turn will into requirements. Prioritizing the asks the AI Office should make of companies, and the risk-modelling methodology to hold them to it (argued in A Call for Better Risk Modelling: this is urgent since CoP enforcement starts on August 2); auditing the thresholds and mitigations companies publish, continuing what AI Lab Watch has been doing (OpenAI’s red line for AI self-improvement is fundamentally flawed); operationalizing and harmonizing red lines across jurisdictions (AI Red Lines: A Research Agenda), or the recent draft treaty proposal signed by a coalition of international experts.

Research that measures the progress in political will. The effect of advocacy is mostly illegible.^[36] But even if the causal chain is hard to be sure of, we can still measure the aggregate effect, and potentially fund more METR-graph-for-policy, like the AIPN tracker cited in Section 1, and better understanding of what’s happening under the hood.

Research that creates demonstrations of risks. Model organisms like agentic misalignment: I use this paper in all my presentations to policymakers now—I think that it is the best paper to demonstrate that frontier models could be dangerously unaligned, which is still one of the main bottlenecks (if not the main one), for AI risk to be taken seriously and prioritized accordingly. I thought in the past that we already had enough risk demonstration (Sleeper agent, alignment faking, Mecha Hitler, ChaosGPT...) but no, this is really a substantial improvement.

Engineering that makes “yes” cheap. Factorize technical mitigations across labs: a shared, off-the-shelf library of safety techniques (constitutional-classifiers-style) that Chinese labs or Mistral could adopt seamlessly.^[37]

Research that could flip the strategy’s sign. Advocacy without red-teaming is how you lock in the wrong ask: Human takeover might be worse than AI takeover. I’d like to fund the research that proves this post wrong and tells me what to do instead. There is no shortage of such cruxes^[38].

Addendum

Two closing notes on why I expect this thesis to become more true over time.

Political will is (or will be) the bottleneck for nearly every cause, not just AI safety

Don’t take all of this personally; AI Governance being slow is not an AI safety-specific concern.

Ok, yes, to some extent, political will being the bottleneck doesn’t apply to causes that are still genuinely researchy with no clear ask, where “keep doing the research” is the right call. But for most causes with a known ask, the constraint is coordination and political will, not more analysis.

I see some people in the ecosystem starting to take space governance seriously, and writing about this. Yet the UN has worked on it for roughly fifty years, and I don’t think the community has started ngaging with the UN on this.^[39]

Same for power concentration: if politicians were AGI-pilled, they would act much more rationally around this.^[40]

(This probably won’t happen in the short term, and it should only be considered after we handle AGI, but getting politicians to be more rational, scale-sensitive, and epistemologically sound in the long term would, in general, be incredibly useful and beneficial for society.)

We can automate the research, but not the consensus

AI will clearly accelerate technical safety work, and even governance inputs. I expect AI to be good at finding ideas at some point and to become excellent at forecasting. But it seems, empirically, that people don’t care about forecasts; they don’t care that top forecasters with a good track record, like the authors of AI-2027, are raising the alarm.

AI won’t, by default, be able to accelerate agreement, consensus-building and human engagement. Consensus is a human process, and it scales differently than research does.

So, as automation reduces the research bottleneck, the human-coordination bottleneck becomes even more important.

...

So yep, that’s bad news for us introverts; we’ll have to engage with the external world at some point.

^
(not disagreement after consideration, but absence of the conversation itself)
^
Update: ~7% engage the broader concept of loss of control, though explicit existential framing stays under 1%. See the discussion in the comment.
^
This isn’t the first time LessWrong has heard the case for reallocating toward advocacy. The post “Instead of technical research, more people should focus on buying time” made a version of it in 2022, and the community’s verdict was that outreach backfires when done poorly, and that indeed it will be done poorly (Wentworth–Larsen), and this was probably reasonable on the evidence of the time; Katja Grace diagnosed the underlying aversion the same year. Then, that verdict has been reopened: MIRI pivoted its entire strategy toward communications, Ruthenis argued in 2025 that awareness is the bottleneck, and gave up on policymakers, which I’ll dispute below. What the reopening has lacked is concrete evidence from inside the rooms. That’s what I try to add: a model of what advocacy produces (the funnel) with 2 years of insider experience across European and multilateral institutions.
^
It would be nice if this law was as solid as the METR’s doubling trend of AI capabilities.
^
Senators Hawley and Blumenthal introduced the AI Risk Evaluation Act, a mandatory pre-deployment evaluation of frontier systems for loss-of-control and scheming behavior, with penalties, and Hawley did it while his own party’s administration was pulling the other way. Bernie Sanders talks about superintelligence and introduced a recent bill. I might be forgetting other initiatives.
^
Chain-of-thought (CoT) is the model’s step-by-step reasoning trace. One of our better safety hopes is that we can read it to catch misbehaviour, but that only works if labs don’t optimise the CoT to look good, which destroys its faithfulness as a signal. Anthropic has acknowledged inadvertently training against the CoT on more than one occasion. Doing it by accident is exactly the kind of basic process failure that should be easy to avoid. See the AI safety Atlas for an explainer on this.
^
Note that I value Greenblatt’s opinion since he has an excellent track record of forecasting AI capabilities.
^
His current AI-2040 numbers put Plan A at ~19% takeover risk vs ~45% for Plan D: a ~58% relative cut rather than ~84% (that requires a near-ideal Plan A).
^
(Note that we don’t need to skip from Plan D to A to get substantial improvement—we can also push continually along the spectrum of political will, where each increment gets a worthwhile improvement)
^
(and people who wield power, which includes some people in AI companies)
^
(at least, if you have the time to read this)
^
Enforcement is a crude lever compared to a request for information and can also introduce adversariality into the relationship with the provider, so it is not entirely surprising to see that the Commission is very careful when using those powers.
^
That’s not surprising considering that even at NeurIPS, the biggest gathering of AI researchers in the world, a third of the AI researchers don’t know what AGI stands for, but yeah, we are nowhere near.
^
I now think that loss of control is probably more urgent than biorisks. See this analysis from PourDemain. I think loss of control is orders of magnitude more likely than irreversible x-risks from AI-enabled pandemics, even if both are probably already at intolerable levels.
^
We matched on exact strings.
^
PS: In the comments, cloud notes that ~7% of submissions raise loss-of-control-type concerns, but loss of control and existential risk differ, and the “<1%” refers to the latter. Reading the submissions flagged under “AI pursuing its own goals,” only around ten frame it in genuinely existential terms (extinction, human disempowerment, uncontrollable ASI), still under 1% of the 1,534. That low explicit signal, even among the actors most predisposed to raise it, is itself the self-censorship dynamic described in the post.
^
When people are asked about AI directly, they often express concern; but when they are asked what political issues matter most, AI risk is usually absent or marginal. Concretely: YouGov’s “AI will negatively affect society” rose from 34% (Dec 2024) to 47% (June 2025), yet AI does not appear at all in Pew’s 24-item ranking of top national problems (Feb 2025) and sits near 1% in Gallup’s open-ended “most important problem.”
^
And the comparison is cross-pollster rather than head-to-head (no single neutral poll ranks them against each other). Pew (Aug 2024) finds people “highly concerned” about misinformation (66%), loss of human connection (57%), job loss (56%) and bias (55%); a separate YouGov poll (June 2025) puts concern about AI-driven human extinction at 43%.
^
Did you know that the Mexican government has already been hacked by an unattributed hacker using Claude? (Bloomberg, Feb 2026) It seems that, in an authorized red-team test, a frontier model reportedly compromised most of the NSA’s classified systems within hours (NYT). The fact that this is not the top story in the world is also revealing about the information ecosystem.
But again, this is not really a factor under our control, and I argue in this post that convincing warning shots are unlikely. I come back to this in the sub-section “The main objection: Let’s just wait for a warning shot? I don’t think this works.”
^
There are more lobbyists working on the EU’s digital files (890 FTE in 2025) than there are MEPs (720). (Corporate Europe Observatory, Big Tech lobby budgets hit record levels)
^
202.5 vs 55.75 FTE, Green-Lowe / CAIP, An Activist View of AI Governance, 2025; author’s estimate.
^
I don’t know the grantmakers’ specific reasons, and they may have been good ones. But I have the feeling that the ecosystem’s revealed preference is that a new research org is easier to fund than an advocacy org.
^
In advertising, the exact threshold of repetition has been debated since Krugman’s ‘Why Three Exposures May Be Enough’ (1972) — two, three, four? — but the general effect is well established.
^
(or at least to reduce this parameter in their weighted factor models)
^
I cannot name them publicly, but DM if you want private proof.
^
The comparison is loose; UNFCCC observers also include industry groups and all kinds of NGOs, while I’m counting only non-technical AI safety organizations (which is also generous in some respects). Regardless, I think that even with heavy discounting, the gap is enormous.
^
Remember also that the environmental movement is over fifty years old. Rome wasn’t built in a day.
^
Michael Dickens’s 2025 donor review also concludes advocacy is far more neglected than research and that the few advocacy orgs do not get much grantmaker support.
^
I’m proud to say that I’ve spent four years on this problem by founding ML4Good (20+ bootcamps in 10 countries). Of ~400 alumni, ~150 now work at places like the EU AI Office, MATS, and UK AISI..
^
The insider game has documented integrity costs, and there is also a missing mood (Integrity in AI Governance and Advocacy). But I also think that it can be done correctly. E.g., I don’t think the most effective strategy is necessarily to open with the full Doom argument, and there are many strategies that lead to effective results and indirectly lead to more political will.
^
Why are Neuralese bad? See this: Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety
^
I’m not sure Europe2031 was net positive. See.
^
Maybe a wild proposal, but one way to bypass institutional stigma is to create safe channels for civil servants, policymakers, and lab insiders to express what they already think privately. Maybe organizing anonymous joint statements?
^
I must now admit that we should have pushed harder to explain the risks during the red lines campaign: removing the detailed explanation from the FAQ was a mistake.
^
The article “diplomacy on a deadline” explains how, historically, we were able to negotiate treaties quickly. Spoiler: feeling a sense of urgency helps tremendously, among other things.
^
“A lot of semi-invisible, ongoing-over-years, and hard-to-definitively-attribute work has gone into many of the policy ‘successes’ of recent years. And sometimes part of the difficulty in attribution is actually down to things like the social proofing of multiple groups providing independent evidence and testimony that adds up to a credible body of expert input on a topic.” – source, Seán Ó hÉigeartaigh
^
It’s been more than a year since the constitutional classifier paper was published, and there is still no ready-to-use library. This is a coordination failure because the same work is duplicated over and over, and because this is currently one of the most effective strategies for reducing misuse. Currently, a small team in each lab might be tasked with reimplementing the best mitigation strategies (which include CC), and safety teams at smaller labs are often just a handful of people. If this library were maintained and updated as new vulnerabilities emerge, my guess is that it would free up a lot of time for safety teams across different labs. This is high-leverage because it would raise the floor for the whole ecosystem.
^
See Zvi’s Crux List for much more than you have ever asked for.
^
More generally, I think that our community has a tendency to ignore the ecosystem and work in silos, disconnected from the institutions and international fora, while that’s where the governance discourse happens.
^
Yes, politics is the mind killer, but only because smart people disengage from it. This is a hyperstition that needs to stop.

How does one plan for death as an effective altruist? by zenc

zenc — Sun, 19 Jul 2026 06:22:50 +0000

I may or may not have dreamt this, but I thought there existed a useful post on the forum about how to write a will as an EA so that your money goes to the most effective charities in the world upon death. For the life of me, I can’t find this post now (pun unintended), so if anyone remembers this post or is able to point folks interested in this question to resources on the subject—I would really appreciate it.

If you’re agentic, work in biosecurity by sharmaayushmaan🔸

sharmaayushmaan🔸 — Sat, 18 Jul 2026 17:38:12 +0000

Disclaimer: Although I work on the Groups Team at CEA, I’m writing this in a personal capacity, and this post does not constitute an endorsement by CEA.

Agency—the realisation that you really can just do things.

TL;DR

Biosecurity needs people (of any background) who are agentic and have a high execution velocity and track record. The field is ready to absorb individuals who feel a heroic responsibility to build teams and start organisations that own key ecosystem gaps. You do not need a background in biology to be a great biosecurity operator. For every stage of your career, there’s an unmet need in the ecosystem you can directly address. If you’re still unsure, I’ll personally pitch you on why and how to best help.

Why GCBRs

There’s been extensive writing on the case for working on existential risks and biosecurity. I signpost some readings here and here that describe this better and more succinctly than I can. I want to focus on why now is the time to commit to biorisk. It’s estimated that around ~160 people are working full-time on reducing the near-infinite ways in which mirror bacteria or engineered pandemics could pose existential risk to humanity’s future and flourishing. Despite this, it’s estimated that almost 95% of this risk can be reduced by deploying relatively tractable interventions: PPE, biohardening, detection, and medical countermeasures (AIxBio is worthy of its own post). It’s incredibly tractable, neglected, and high-impact to work in a field where there’s such strong consensus on how to reduce risk. The primary constraint at the moment is people.

It’s an exciting time for the ecosystem

There are strong reasons to be excited about biosecurity fieldbuilding right now. Cambridge Biosecurity Hub has had success growing the community and becoming an owner for stewarding the movement; Bluedot is trialling the Voyager Fellowship to locate much-needed entrepreneurial talent after relaunching the Biosecurity Course; Pivotal has launched a Biodefense Fellowship; CG and Bluedot have begun collaborating on rapid and career transition grants. Personally, many of my collaborators and friends have transitioned into direct work in both the GCBR and meta-space, where the possibilities for introducing biosecurity to talented people are growing.

AI Safety has recently had a generalist resurgence that biosecurity could draw learnings from in how we characterise talent gaps. There has also been exciting growth within EA/AIS ecosystem building (e.g. Kairos’ SPAR and CBAI starting AIxBio fellowships/streams; the launch of Impact First; NEST, and growth within CEA) that I’m excited to see more collaboration and cross-learning between. Few spaces are as welcoming of entrepreneurial and fast-shipping talent as biosecurity right now.

If you’re still a student

I spend most of my time at CEA liaising with students looking to transition into direct GCR work. Although the GCBR community tends to be geared towards mid-career/senior professionals, there are still ways to contribute whilst upskilling. Students are in an enviable position of limitless time and energy, which we should all be leveraging more.

Hosting events
1. Full-time staff have limited capacity to do this. Hosting happy hours, offering support to those trying to arrange summits, and working with your universities to host convenings with academics in adjacent fields (e.g. metagenomics, supply chains, etc.) can be unusually high-tailed.
Launch a biosecurity group on your campus (especially if you’re at a top university)
1. EA and AIS have grown considerably due to the bottom-up efforts of local groups. Students (from undergraduates to postdocs) who lead effective biosecurity groups can provide an early-career funnel for organisations looking to hire junior talent, and most importantly, make biorisk one of the top issues students and graduates want to work on. A single group could steer dozens of students into the field.
  1. I’ve become increasingly sceptical of concerns about the delayed impact of students. I think there’s far more to do than just direct technical work, and generally increasing the worldwide awareness of biorisk seems counterfactually worthwhile. At the same time, student AIS groups such as OASI, MAIA, and AISST have provided a model for student involvement towards GCRs I’d be curious to explore in biosecurity.
2. Generally speaking, growing your local EA group can increase the peripheral knowledge about GCBRs in your school and the agency of students to pursue an impact-aligned career.
Encourage career transitions
1. Students often overlook this. Agentic community builders (especially students) have privileged access to coffee- and career-chats with those later on in their careers. Biosecurity needs individuals with a broad skillset across a range of fields (from operations, logistics, entrepreneurs, civil servants). Pitch people on why they should work on biorisk, and encourage them to do a BlueDot course or fellowship. A counterfactual career-career transition contributes thousands of person-hours of additional capacity if sufficiently scoped.
Experiment with media communication
1. I want to caution that, in this context, I encourage students to write for their college papers and outlets about:
  1. Topics they feel are neglected and that people should work on (e.g., research narratives about mirror bacteria)
  2. Fellowships or opportunities in GCBRs
  3. Encouraging students to be agentic and avoid common failure modes of high-performing talent

If you’re early/mid-career

Although I have less interface with mid-career professionals, there are a multitude of ways to get involved in biosecurity right now. The ecosystem needs people with expertise that is useful to GCBRs, and it’s widely understood that individuals can pick up the necessary context. We need creative problem solvers who are ready to use their background towards one of the four pillars in biosecurity.

Skills I’m especially excited about include:

Field-builders (of all stages) looking to collaborate on ecosystem growth
People who have experience working with the government, national security, or policy
People who have ops/chief of staff/founder skillsets
People who have worked in supply chains, logistics, or distribution are taking on PPE stockpiling
Seeing the more traditional AIS crowd consider taking on AIxBio challenges

I list below a first-pass list of ways to get involved. Much of this is also captured on https://biosecurity.world:

Courses

Biosecurity (BlueDot)
Biosecurity Fundamentals (Cambridge Biosecurity Hub)

Fellowships

Frontier Biodefense Fellowship (Pivotal)
AIxBiosecurity Fellowship (CBAI)
AIxBiosecurity Fellowship (ERA)
SPAR (Kairos)
Horizon Fellowship (Horizon Institute for Public Service)
Emerging Leaders in Biosecurity (Johns Hopkins Centre for Health Security)
Fellowship for Ending Bioweapons Programs (Council on Strategic Risks)
Mid-Career Biodefense Bootcamp (Council on Strategic Risks)

Projects to own

Places to build

Voyager Fellowship (BlueDot)
Incubator Week (BlueDot)

RFPs and funders

RFP on Biosecurity (Coefficient Giving)
Career Development and Transition Funding (Coefficient Giving)
Intercept
Pathogen-Agnostic Threat Detection RFP (Blueprint Biosecurity)
Funding opportunities (Blueprint Biosecurity)
Bioresilience Fund (Blueprint Biosecurity)
Rapid Grants (BlueDot)
Long-Term Future Fund (EA Funds)
Manifund

To follow

If you currently work at a biosecurity org

While most fieldbuilding discourse is centred around vertical movement through the ecosystem, I’m really excited to see increased communication between fieldbuilders and those working in direct (technical or policy) work. One thing I’ve noticed about AIS is the shared sense of responsibility towards ecosystem growth by those working in organisations of all sizes. As a fieldbuilder, I’d encourage those currently working to consider where their comparative advantage is in contributing to the mission.

Encouraging more thought-leadership/public-facing writing in your organisation
1. Notwithstanding the sensitive and infohazardous nature of a significant portion of GCBR work, writing shallow articles on new focus areas, topics you think should be worked on, or where you would like to see the field can help fieldbuilders better prioritise our referrals, connections, and initial conversations. I’ve found the SecureBio and Defenses in Depth substacks particularly useful. I do believe the EA Forum is still a strong place for the community to convene and exchange ideas.
  1. This could come in various forms, from posting on Linkedin, writing on Substack, or engaging in forum-posting.
Mentor (early-career) talent
1. Having launched and scaled almost 3 mentored-research programs thus far, the biggest constraint for any program is mentor availability. The value-add of your time is highly asymmetric—even 20-30 minutes a week could be the deciding factor in a counterfactual career transition. Many mentors I have worked with tend not to consider this as an option until asked; I strongly encourage anyone (at any stage) to consider becoming a SPAR mentor, support a fellowship, or even post a project on the forum seeking mentees. Effective Thesis has a bank of research topics that’s routinely updated to draw and contribute to.
Launch your own programs
1. Consider whether it makes sense to launch your own research, fellowship, or entrepreneurship programs. It’s undeniable that there is a staff cost to this, but the upside of well-scoped programming that meets specific needs could be huge. Take the example of AIxBio and the successes of fellowships in directing the talent pool towards this particular area. I’m keen to speak to orgs considering their own fellowship programs/help soundboard.
  1. An alternative here is collaborating to start local biosecurity hubs, a shout-out to Li-Lian Ang, who wrote about this for AIS.
Tap into the meta-space
1. The meta-space has unrealised potential and reflexivity to wider EA/GCR cause prioritisation. I’d be excited to see those working in biosecurity tap more into the institutional knowledge of GCR and EA capacity builders. In our roles, we’re constantly strategising about community growth, funnel conversion, and how to increase the pool of talented (EA- or impact-aligned) individuals. Growing biosecurity is as much of a cause-agnostic field-building challenge as it is a biosecurity-specific one.
  1. I’m fairly convinced that there’s a greater institutional framework for developing and directing talent than the community may appreciate. I suspect the limiting factor here is a wider coordinated fieldbuilding effort/strategy.

(Additional) Disclaimer

Many of these ideas demand far more nuance than I’ve given. I strongly encourage readers to red-team these ideas and treat them as prompts for your own reflection and discussion.

I’d be really excited to see a forum discussion on:

Ways to better coordinate as a field
What thought-leadership in biosecurity fieldbuilding should look like
Exciting fieldbuilding projects people are considering or looking to collaborate on

Europe 2031: A Response, and Call for Open-Ended Inquiry by Sarah Weiler

Sarah Weiler — Sat, 18 Jul 2026 15:22:50 +0000

Tl;dr

This is a response to Europe 2031, a scenario published recently with significant resonance across the European policy ecosystem.
Europe 2031 is a scenario-driven argument: In order to avoid being powerless vis-a-vis US and Chinese geopolitical and geoeconomic coercion, Europe should heavily invest in data centre construction on European territory. Having data centres to run advanced AI models preserves European autonomy in an age where AI drives productivity and economic competitiveness. (more below, or just head to the authors’ own summary^[1])
My response complains that this argument rests on assumptions^[2] which are left implicit in the narrative, when instead they should be made transparent. There is much that we do not understand well about how Europe can respond to accelerating AI development and power concentration, and we need open discussion about those uncertainties and competing responses.
I respect the authors of Europe 2031 for their genuine effort to wake up European policymakers to a forward-looking response to advanced AI development. But I also think that publications like Europe 2031 are net negative in their effect on the world if they foreclose discussion rather than encouraging it. This post is an attempt to push further in the encouraging-discussion direction. It’s meant as a complementary effort to what the scenario is trying to do, not an adversarial one.
In that spirit, here is the overarching question I would love to see discussed more^[3], rather than less, amongst all those who engage with Europe 2031’s call for more serious EU AI policy endeavours:
How do we overcome the coordination problem of global AI governance? How do we prevent extreme power concentration within a few leading companies (and individual company CEOs)? What role can Europe play in responding to these challenges?
Also in that spirit: I’m listing a few resources that have started these discussions in this GoogleSheet: EU_AI_geopolitics_resources. If you know of or encounter others, please do drop them in the comments or submit them via this form, and I’ll endeavour to keep the sheet a living resource repository going forwards.

Context and epistemic status

This is a brief essay about my concerns regarding Europe 2031 and the way it may be read and received. I was a bit reluctant to write this because I understand and share the authors’ frustration with European inaction, which is in part fueled by internal division and endless bickering. But I’m also reluctant to not write down what worries me: the publication of Europe 2031 may have net negative consequences if we don’t receive it critically and if we use it as an end rather than a start to discussions around the EU’s geopolitical AI strategy. I hope that this essay helps us think more critically and intentionally about the futures we envision and about how we work towards them. It is very tentative, not presenting final solutions but raising questions and prompting us to look more carefully before locking in onto any one strategy. I mean for this to be a critical but constructive contribution to the effort Europe 2031 has started, not an adversarial attack.

I spent roughly 3-4h drafting, discussing with friends, and formatting the essay below. It could be much better, more thoughtful, and more incisive in the arguments it is making (especially in its identification of assumptions in Europe 2031 and its outline of possible alternative perspectives). I am keen for criticism, refinements, and expansions of the ideas I start to develop below; as I emphasise above: the goal is engaged discussion, including discussion that takes apart what I’m saying!

The actual post

Intro and main argument

Before launching into my critique, let me note what I appreciate the scenario for: it points at what I feel are valid and needed concerns about EU officials and institutions being indifferent to AI and its possible geopolitical implications, about the incredible lethargy of European bureaucracy, and about people with authority unjustly dismissing thoughtful input from lower-ranked colleagues. I share a sense of frustration with all of these, as well as a desire for things to improve. And I think Europe 2031 may be an effective prompt to bring some such improvements about, which I feel grateful for.

Why, then, do I worry about net-negative outcomes from the publication of the scenario? Because I think that it unreasonably presents one kind of AI risk and one bucket of solutions to mitigate that risk, making the implicit case that this is the one risk we need to be most concerned about and hence the one bucket of solutions we need to push for. That implicit case is—or should be—deeply controversial, based not on solid evidence but on a bunch of value judgement and empirical hunches. Making such value judgements and empirical guesstimates is fine; inevitable, even, if one is to venture into the space of policy recommendations at all. But presenting them hidden underneath a well-written and captivating story, rather than making the case openly and explicitly and with epistemic humility and transparency where warranted, is not inevitable.

My reading of what Europe 2031 is telling us

The one risk that is presented in the scenario runs roughly as follows: frontier AI is presently being developed mainly by a few US companies, with Chinese competitors reasonably close behind whilst European-developed models remain far from the frontier. That makes European end users dependent on access to foreign AI models. As AI integration starts to dominate growth and productivity potentials around the world, dependence on foreign AI models makes Europe a vassal of whoever it can rent these models from. Such dependence basically disenfranchises Europe as a sovereign entity because the providers of frontier AI models can derive crushing leverage from Europe’s need to deploy frontier models. A disenfranchised Europe is a Europe without a viable future, and certainly one that is incapable of defending its values and European citizens’ interests in the decades to come.

The bucket of solutions, if I read the scenario correctly, mainly revolves around the building of compute capacity on European territory: data centres primarily, possibly also a greater portion of the supply chains needed to build the semiconductors that go into those data centres, with the goal of providing model access and model development capacities through European infrastructure.

A critique of what Europe 2031 is telling us (and of how it tells us)

Europe 2031 tells a story that is not without merit, nor one built on demonstrably false premises. But it is a bit arbitrary: other stories would be possible that would seem equally meritorious and would equally rest on not-demonstrably-false premises. Alternative stories could, for instance, depict a dystopia where a massive European data centre build-out sacrifices regional standards and distracts from global governance efforts to slow down AI development, while at the same time doing little to reign in the power of the United States (which continues to house leading frontier AI companies) over Europe. Or, to pick one other among many possibilities, such a story could portray a future where access to the newest AI models is only marginally important because progress in frontier models easily leaks out to help improve replica models (similar to what we seem to be seeing with Chinese model development at the moment), which means that industrial productivity, innovation, and research capabilities are nowhere near as badly affected by US frontier model export controls as Europe 2031 insinuates.

The choice of which of these stories to tell is arbitrary rather than random or coincidental because there is a specific worldview that underlies the narrative adopted by the authors of Europe 2031: one that privileges deregulation’s short-term gains for innovation over its longer-term risks to societal and environmental sustainability; one that accepts competition as an inevitability and holds that collective action problems, at least at a global scale, simply cannot be overcome, no matter the costs of failing to overcome them; and one that has accepted the risks of rapidly evolving AI development and thus absolves even self-avowedly responsible actors from the burden of fighting such recklessly fast development.

Building more data centres with fewer regulatory hurdles in the EU is not the one sensible response to our emerging AI-shaped geopolitics. It is a heavily assumption-laden response: It assumes that more data centres really will make the EU more autonomous and immune to US coercion, even though leading AI companies will continue to sit in the US (thus giving the US leverage even if it does not have a near-monopoly on compute). It assumes that more data centres in the EU will not speed up capability development, or that sped-up capability development is not a risk worth fighting (despite dire scenarios about the possible ramifications of AI models that improve faster than society can adapt and faster than safety researchers can build guardrails). It assumes that material assets are the only real instrument of power, the only truly secure option to protect the EU’s sovereignty and global influence. And it assumes that the conflict exists between countries (or country blocks), rather than between power-amassing monopoly companies and everybody else.

I can’t help but feel that following the recommendations laid out by the authors of Europe 2031 is giving away much of what we should be fighting for: a chance to mitigate extreme risks of AI; a chance of preventing extreme power concentration in the hands of a few recklessly accelerationist companies; and a chance to think critically about pathways to AI development where technical progress does not outpace the wisdom and resilience we need for a future where technology actually serves flourishing, rather than dooming it.

A plea for rigorous and honest discussion

In other words, the questions I’d love to see contested and discussed, in cooperative but non-groupthink-y ways:

What exactly is the pathway through which increased computing capacity ensures European sovereignty? What are the assumptions required for that pathway to work, and how uncertain are those assumptions?
Does increased demand for data centre production influence the speed of AI capability improvements?
What are alternative strategies for advancing European values and interests in an age of advanced AI development?
- How can the EU play a role in reducing the catastrophic risks from unsafe model development?
- How can the EU reduce the systemic risks of AI integration in its economy and society?
- What level/type of regulation and public action is required to implement these strategies?
More broadly: How do we overcome the coordination problem of global AI governance? How do we prevent extreme power concentration with a few leading companies (and individual company CEOs)? What role can Europe play in responding to these challenges?

A living repository, and continued discussion?

In the spirit of wanting to spark and advance discussion, I’m collecting a few resources that speak to the questions and arguments above: EU_AI_geopolitics_resources. If you know of or encounter others, please do drop them in the comments or submit them via this form: https://forms.gle/1tW1VAE9tA9QrxSKA. I’ll endeavour to keep this a living resource repository going forwards.

***

Lastly, as a PhD researcher studying the epistemics of foreign policy in a context of inconclusive evidence, I’m always keen to meet and convene policy analysts working on some of the most tricky knowledge problems we face in trying to tackle global challenges. If you are one of those, and might be interested in discussion exercises or joint projects with other researchers trying to make sense of Europe’s role in global AI governance, do reach out: sarah.weiler@strath.ac.uk.

^
For a summary of the scenario itself, focus on the first few sections. The last section of the linked summary outlines a range of policy recommendations, some of which go beyond what is discussed in the scenario itself (as far as I recall/understood). That last subsection fits neatly with my call for transparent argumentation and broad discussion. My argument in this essay is that this transparency and honest argumentation is lost in the scenario itself.
^
This post is a quick write-up, so no claim to exhaustiveness, but I identify some of the assumptions that strike me as particularly contentious and debate-worthy below.
^
A longer list of (sub-)questions in the concluding section below.

Locking myself into a room until I donate €1M. by Andrew Neiman🔸

Andrew Neiman🔸 — Sat, 18 Jul 2026 14:27:03 +0000

Disclaimer: While the ideas and content are my own, I used AI to improve the writing, style, and grammar.

Can relentless obsession turn into €1,000,000 worth of good?

That is the question I am determined to answer for myself.

Until I reach my goal of earning and donating €1,000,000, I will, figuratively speaking, lock myself in my room and dedicate nearly all of my waking hours to my full-time job and building my “founding to give” business alongside it.

I must admit that this is not an original idea. I was inspired by this YouTube video about an entrepreneur who adopted an extremely regimented lifestyle while working toward his goal of earning one million dollars.

I want to apply that same intensity to an objective I find more meaningful: not merely earning €1,000,000, but also giving it away to the most effective charities.

What I mean by locking myself into a room is defined by the following rules:

1) Work 12 hours a day, every day, divided between my full-time job and my founding-to-give business. There are no weekends or scheduled days off.

2) Make the schedule sustainable by sleeping eight hours every night. When working on the business, I work in blocks of 90 minutes, with 30 minutes of rest between each block. Rest can include walking, meditating, journaling, light exercise, or listening to music.

3) Cut out all distractions: no social media, YouTube, movies, television or gaming.

4) Retain only enough money to cover basic food, shelter and essential living expenses. All remaining money will either be reinvested into the business to increase its future earning potential or donated to effective charities.

5) Continue following these rules until I have actually donated €1 million to effective charities.

Why am I doing this?

Because I believe this may be my best bet at doing the most good.

While searching for a high-impact career, none of the direct-work opportunities I encountered spoke to me or felt like a particularly strong fit. I found myself much more drawn to earning to give.

My first plan was to pursue quantitative trading, a field known for both its high earning potential and its highly competitive hiring process. However, after two separate rounds of applications and being rejected everywhere, my chances of landing a quant job seemed close to zero.

During that period, I landed a different job in finance and came up with a business idea.

I truly believe in this business and currently consider it my best personal fit to creating a significant amount of positive impact.

Of course, believing in an idea and working hard do not guarantee success. There are countless variables that determine whether a business succeeds or fails, and most businesses fail. I understand that mine has a high chance of failing as well.

But I would rather make a serious attempt and fail than spend the rest of my life wondering what might have happened if I had fully committed myself.

A personal experiment for myself, not a guideline for others.

I want to make it very clear that I do not necessarily believe this lifestyle will lead to the greatest impact for everyone, nor am I advocating that others adopt it.

There are two main reasons for this.

First, there is much more to life than working and trying to maximize impact. I hope everyone in this community can have a healthy and happy life alongside the impact they create. Finding the right balance between doing good and living a fulfilling personal life is a deeply personal question.

Second, more hours worked do not necessarily lead to greater impact. An additional hour of relaxation or socializing may improve focus, make you more efficient or prevent burnout.

Okay, so then why am I really doing this?

I wish I could say that the answer is something heroic, that I am sacrificing everything for the greater good of humanity.

But the truth is that, for me, this is not entirely a sacrifice. It is something I would probably be drawn to doing anyway.

I have always had an obsessive personality.

When I first discovered chess, I spent my days doing almost nothing else. When I was not playing matches, I was reviewing my games, solving puzzles, or watching educational videos by chess grandmasters.

When I first became interested in fitness, I worked out every day, tracked every calorie and gram of protein, tried very hard to create the perfect workout routine to maximize muscle gain and spent much of my remaining time watching videos of other people vlogging their workouts.

Obsession can be unhealthy, unproductive, or directed toward things that ultimately do not matter. I do not want to glorify it. I simply want to figure out if I can point that part of myself toward something more meaningful.

Instead of obsessing over a chess rating, a physique, or another purely personal goal, I want to obsess over creating as much positive impact as I can. That, to me, is a life well spent.

Thanks for taking the time to read this. Now if you’ll excuse me I’ve got some work to do.

Planetary Defence (PD) - the most complete effort to address a cosmic threat to date: a short review by MaxSJC

MaxSJC — Sat, 18 Jul 2026 12:20:08 +0000

Planetary defence—the effort to detect, track and intercept asteroids that might impact the Earth—is the most complete (and only?) effort to tackle a cosmic threat (asteroids/comets, though way more asteroids than comets) by far. Unlike most cosmic threats which remain a topic of speculation, planetary defence has:

Policy support:

In contrast to the funding cuts in the FY2026 President’s Budget Request for NASA, PD funding actually increases.
A number of PD-dedicated organisations such as the International Asteroid Warning Network (IAWN) and the Space Mission Planning Advisory Group (SMPAG) has been set up to quickly assess and mitigate any emerging asteroid threat. (IAWN operations was first put to the test by the asteroid 2024YR4 in late 2024 - early 2025.)

Scientific effort:

Numerous current astronomical surveys (which scans large parts of sky to detect and study objects) are dedicated to find new asteroids and track already discovered ones. Upcoming surveys such as the Vera C. Rubin Observatory (just started its operations!) and NEO Surveyor (a main reason for the FY2026 budget increase) is expected to expand the effort drastically.
The Double Asteroid Redirection Test (DART) was launched in 2022 to test out a method to deflect asteroids and the results from the mission is still being analysed till this day—in fact, the ESA HERA mission is expected to do follow-up observations later this year!

This resulted in huge progress made in characterising the threat from asteroid impacts and developing interventions. So far:

over 90% of the asteroid which poses existential threats (size larger than 1 km) has been found
we launched multiple probe missions to asteroids, studying their structure and composition: asteroid properties are important in knowing the effectiveness of interventions!
we have a tested intervention to stop an asteroid of medium size (hundreds of meters)
we tested our ‘action plan for a potentially harmful asteroid’ on the asteroid 2024YR4: It is discovered on 27 December 2024. Multiple observations are made with major observatories in the coming weeks and months. We ruled out the impact possibility by March 2025.

Future work seems to lie in these directions:

finding at least 90% of asteroid of sizes over 140 m: these can deal at least regional damage
continue tracking found asteroids: they need to be continuously tracked over centuries, we still cannot predict their long-term orbits very well!

However, current PD efforts are not focused on comets. I have not gained much understanding so far but the biggest problem with comets so far is that unlike asteroids, they come from the outer regions of the solar system, which means:

they come into Earth’s orbit at a higher velocity, meaning we can only detect a harmful comet a short time before impact
they do not do multiple orbits within a human lifetime, making it hard (or perhaps useless) to determine their orbits
they consist of volatile substances which can only form far from the Sun; as they get closer to the Sun, these substances boils and outgas, pushing the comet like a rocket, making their orbits even harder to determine!

Seems like the only way to deal with a harmful comet is to find it as early as possible and come up with interventions as it comes towards us. Thus the bottleneck seems to be detection capabilities and perhaps a further understanding of their near-Sun behaviour.

SUMMARY

the effort to mitigate risks from asteroid impacts are well-funded and attended unlike other cosmic threats
we have identified most of the extinction-causing asteroids, now working on intervention development, continuous tracking and identifying asteroids that cause at least regional damage
the effort to mitigate risks from comets remains an open question, perhaps with detection capabilities most needed

A proposal for food retail and services: the internal animal welfare feebate by Stijn Bruers 🔸

Stijn Bruers 🔸 — Sat, 18 Jul 2026 00:03:20 +0000

Most consumers value animal welfare. According to most people, the cost of animal suffering in the production of animal-based food, is huge. In monetary terms, the suffering of most broiler chickens would correspond to 8000 euros per kilogram of chicken meat. This external cost of animal suffering is much higher than the external costs of e.g. environmental impacts, public health risks or climate change. Of course, an animal welfare levy of 8000 euros per kilogram chicken meat would immediately wipe out the entire meat production sector, and would not be politically feasible. There is a related but much more feasible proposal for the food retail and services sectors (supermarkets, restaurants,…), that is likely to have a wide support: an internal animal welfare feebate.

A feebate is a combination of a fee and a rebate. In the proposal, the fee is an animal welfare levy or tax on products that involve animal suffering, such as meat, and the rebate is a subsidy for animal-free (plant-based) alternatives. An internal feebate means that the feebate is applied not by a government at the national level, but internally by a food retail or service business at the company level. Supermarkets or restaurants sell animal products at a price higher than the production costs and use the extra revenue to cover the losses of selling plant-based meat alternatives at a price below the production costs.

In more detail, the most feasible proposal consists of a 2 to 3 euro per kilogram levy on products from animals smaller than a pig (in particular chicken meat, eggs, fish and shrimp), and a subsidy on the best (tastiest, healthiest) plant-based meat alternatives (in particular plant-based chicken fillet and nuggets), with a subsidy cap such that the meat alternatives do not become cheaper than their untaxed animal-based counterparts (i.e. the price of the plant-based alternatives cannot fall below the current, baseline price of chicken meat).

This proposal should be feasible, because:

according to a survey (in Belgium), the majority of people are willing to pay a 3 eur/kg animal welfare levy on meat,
according to large blind taste tests, a majority of consumers think that some plant-based chicken(less) fillet and nuggets taste at least as good as their animal-based counterparts (chicken fillet and chicken nuggets),
most of the subsidized plant-based chicken meat alternatives have a similar or better nutritional value than chicken meat: they contain essential protein, minerals and vitamins as in chicken meat, and additionally contain more healthy fibres than chicken meat),
according to a rough calculation (see addendum below), the revenue from a 2€/kg welfare levy on chicken meat can probably finance a subsidy for plant-based fillet and nuggets such that those plant-based products can be sold at the current price of (untaxed) chicken meat, which means that consumers keep the option to buy a product that is at least as tasty, healthy and cheap as current chicken meat,
the feebate can be applied at the level of a company and doesn’t require a new tax law, nor extra governmental administration and control,
only the animal products that have by far the highest welfare footprints (i.e. the most animal suffering) are taxed,
the feebate has additional environmental benefits (plant-based meat alternatives have a lower environmental footprint than animal-based counterparts) and public health benefits (plant-based meat alternatives do not increase the risk of infectious zoonotic diseases), such that the feebate fits within corporate social responsibility and Environmental, Social and Governance (ESG) policy,
the subsidy cap ensures that the system is stable (without the cap, the price of subsidized plant-based alternatives may become zero, consumers may completely shift towards the free plant-based alternatives, the animal-based products market may completely collapse and hence there will no longer be any tax revenues to finance the subsidy).

With this proposal, the quantity of chicken meat could decrease with about a quarter. As most farmed animal suffering is due to the suffering of chickens, this proposal could potentially decrease farmed animal suffering with more than 20%.

Addendum

To measure the economic impact of a feebate on chicken meat, we can start with a consumer’s utility function with a constant elasticity of substitution (CES):

with

the consumed quantity of meat, in particular chicken fillet,
the consumed quantity of plant-based meat alternatives, in particular chicken(less) fillet,
the elasticity of substitution (due to lack of estimates in the economics literature, I assume this parameter to be equal to 2 for ease of calculations),
the price elasticity of demand (assumed to be equal to −0,7, according to a systematic review),
and
the non-monetary preferences for resp. chicken meat and alternatives, representing taste preferences, food culture, cooking habits and familiarity.

The current prices for chicken fillet and alternatives are respectively

and

eur/kg (according to average retail prices of animal-based and plant-based chicken fillet in Belgium, such as the plant-based fillets of The Vegetarian Butcher and Vivera that scored highest on the NECTAR taste test).

Introduce an animal welfare tax

on animal-based meat that finances a subsidy

for plant-based alternatives, such that the new equilibrium prices become

and

As the subsidy is financed by the tax, we have the budget constraint:

Here,

and

are the equilibrium quantities of chicken fillet and plant-based alternatives in the presence of a tax. In the absence of a tax, we can use volume units such that

for simplicity. As in the current market (without a feebate), plant-based fillet covers around 1% of the market for fillet, I assume

. With these values, and the equation

the non-monetary preferences can be calculated to be

and

. This means that apart from price, consumers have a 15 times higher preference for animal-based chicken fillet than plant-based fillet.

To calculate how much chicken meat is reduced due to the feebate, we can express the equilibrium quantity as

with

the aggregated price index in the feebate system and

the aggregated price index without the feebate. This equilibrium quantity has two factors. The first factor (with the price elasticity of demand

) represents the total market shrinkage effect (i.e. a decrease in the total consumption of animal-based and plant-based fillet due to an increase in aggregate price). The second factor (with the elasticity of substitution

) represents the market shift effect (from animal-based fillet to plant-based fillet) due to a change in prices.

There are two operational phases:

the budget-balanced phase whereby the tax revenue matches the total subsidy payout,
the fiscal surplus phase, when the tax revenue is higher than the total subsidy.

The threshold between the two phases lies at a tax rate of 1,17 eur/kg.

In the fiscal surplus phase, the excess tax revenue can be used to subsidize the price of other plant-based alternatives of animal products. The table below summarizes the results.

Tax rate (eur/kg)	Chicken fillet price (eur)	Alternative plant-based fillet price (eur)	Chicken fillet quantity bought relative to baseline	System phase
0	9	23	100%	Baseline (no policy)
0,5	9,5	16,2	95%	Budget balanced
1,17	10,17	9	87%	Cap threshold
2	11	9	79%	Fiscal surplus
3	12	9	70%	Fiscal surplus

At a tax rate of 2 eur/kg, the quantity of chicken fillet decreases with 21% and the plant-based fillet price equals the baseline chicken fillet price.

Note that in this example, the plant-based alternative (fillet that tastes as least as good as chicken fillet) has a very high baseline price: 250% times the price of chicken fillet. Even with such a large price gap between the animal-based and plant-based products, a relatively moderate tax rate of 2 eur/kg could drop the price of the subsidized alternative to the baseline (untaxed) price of the animal product.

When the elasticity of substitution becomes higher than 2,5, or the current market share of plant-based fillet would be higher than 2%, the threshold tax rate would be above 3 eur/kg. In that case, a tax rate below 3 eur/kg would not suffice to drop the price of the plant-based product to the baseline price of the animal product. As both the prices of the animal product and its plant-based alternative are higher than the untaxed price of the animal product, consumers are no longer able to by a product that is as cheap as the current animal product.

EAs Should be Vegan, for EA reasons by William Harrison 🔸

William Harrison 🔸 — Fri, 17 Jul 2026 22:28:20 +0000

This post is, to a small extent, a response to this post by Lincolnq about taboo-ing the “V word.”^[1] It is also a response to a general attitude I see in EA spaces that is dismissive of veganism as a practice and seems inclined to prefer donation, even in very small amounts. Mostly, it’s a case that being vegan makes you a more effective altruist.

None of this post rests directly on the claim that veganism actually reduces the number of animals raised, tortured, and killed in expectation. I propose 5 main benefits of veganism to EAs: better moral reasoning, more credibility for non-EAs, better signaling about small animals, a constant sense of purpose and motivation in small acts of abstinence, and veganism’s abolitionist praxis.

Note: I’m a hypocrite, and am not quite vegan.

Cognitive Dissonance and Motivated Moral Reasoning

As much as we pride ourselves on checking our own moral and cognitive biases, we would be bold to claim that we are completely successful. If you, like me, are complicit in the animal torture apparatus that is factory farming, you might be underweighting the suffering of certain animals or classes of animals, since it feels uncomfortable to acknowledge the depth of the pain you are causing. You might also be spending less time thinking about animal issues for the same reason.

I am a strict vegetarian and am in the process of shifting towards veganism. Even still, I notice myself thinking less than proportionally about dairy cows and egg-laying hens. I do my best to fix this, but their suffering is more uncomfortable for me to ponder than that of shrimps and pigs, and so they are underrepresented in my passive stream of animal welfare thoughts.

To me, one of EAs biggest appeals is its insistence on a moral obligation to discern truth and act on it. Going vegan is an act of epistemic humility that is aligned with that obligation; in doing so, we recognize that we are never fully capable of separating our moral thoughts from our feelings and day-to-day practices (and that, maybe, we shouldn’t always try to).

Credibility to Non-EAs

It is confusing to non-EAs when an EA says that factory farming is among the biggest moral atrocities but continues to participate in it. It seems hypocritical, diminishing the individual EA’s perceived moral credibility and thereby the credibility of EA as a whole. We should be careful to avoid the suggestion that, to participate in effective animal advocacy, one must be vegan, but we should not go about this by ditching veganism altogether. We can maintain our moral credibility without creating a barrier to entry simply by letting newcomers know that we blame systems, not people, and are welcoming to animal activists who work at all levels, dietary and non-dietary alike. Veganism can still be encouraged.

Signaling Expanded Moral Circles

Half-measures leave non-EA onlookers with the impression that only certain animals matter. It occasionally comes up in conversation that I am vegetarian. Many people have responded to this with some variant on “Oh cool! I’m pescetarian!” or “Great! I don’t eat mammals.” I have also gotten responses about adopting shelter animals.

I have an inkling that people group vegetarianism, pescetarianism, avoiding beef, and maybe eating insects together, and view veganism as a separate category that sees all animals as having moral value. “Animal activism” surfaces images of the better funded and more culturally salient companion animal shelters. Veganism is, by nature, all-encompassing, and leaves no room for exceptions or unique emphasis on individual species. Being vegan, thus, conveys the breadth of our moral priorities and leaves non-EAs less room to align their view of us with their own ideologies and diets.^[2]

I will admit that this is my weakest point, partly since veganism is now associated with health trends and environmental movements. Even if that is the case, though, having a recognizable name is crucial for us to curate veganism’s image as a moral movement, not a health fad. I’ll borrow from Lincolnq’s point about the importance of the term “sexual harassment” in naming and then preventing the behavior. We must have a recognizable term to build a movement behind, and it must be a movement that opposes all animal cruelty, especially to inverterbates and small animals.

Lincolnq brings up the Ridglan Rescue to argue that “there are lots of pro-animal people who eat meat.” I think this is a category error. Many people are pro-dog and vehemently oppose cruelty to them. Very few of them carry that same energy to opposing factory farming. Advocacy for dogs is not an instance of the broad moral circles EA aims to cultivate. Veganism is.

Persistent Reminder to Consider Nonhuman Animals

If you need such a reminder, checking ingredient labels when making purchases makes it difficult not think about animal welfare on a daily basis. It’s a constant reminder of the scale of factory farming and the wide reach of animal products. It is also motivating; every such decision reaffirms your own commitment, and is a tangible bit of change that builds your sense of purpose and momentum.

Abolitionism

Most prominent EA animal organizations are welfarist by design. It is at times demoralizing that one of the crown achievements of EA animal welfare is the most effective institution in history at making shrimps die a little less consciously, probably. I say this not to disparage SWP. I think they do amazing work. Welfarism is great and, in my view, certainly outperforms abolitionism on the margin, but welfare progress alone will never lead to the end of factory farming, and will likely not get it to a state of net moral positivity (at least for the animals in the farms). Veganism is abolitionist by design and steers the movement toward the aim of eventual abolition, and is thus a more motivating project for its current- and would-be activists.

^
I understand this is not a complete, line by line response to that post, and I wasn’t aiming to do that.
^
I will grant that veganism, often interpreted deontologically, is not always well-suited to handling tradeoffs and prioritization. Still, with veganism defining our aims, EAs can debate how to order our steps towards a less cruel world