Longview Philanthropy is looking to fund proposals to better understand and reduce the risk of AI enabling the extreme concentration of power — the possibility that AI will enable a small group of people to gain durable control over the lives of everyone else.

We are funding work across two tracks:

• Grants for projects, programs, new teams, and new organizations across twelve priority areas (listed below). We expect typical grants to initially be between $100K and $2M/yr, covering 6 to 18 months, but we will consider proposals outside this range when the work justifies it, and larger renewals are possible once there is evidence of success.
• Career funding for individuals developing or transitioning skills into the area, including course buy-outs, post-docs, training, independent study, transition stipends, and relevant expenses.

You can apply for Grants here, and Career funding here. The application deadline for both tracks is July 2nd 2026, though we’ll assess proposals on a rolling basis as they come in.

For questions, please email power-rfp@longview.org. For information about Longview Philanthropy, see here.

This may be one of the world’s most important problems. If you can contribute to one of the below areas, you may be able to help a vast number of people. We hope you apply.

---

Table of Contents

---

The Problem

We think that advances in frontier AI systems in the next five years could confer enormous power to those who control them. They could gain strong influence over:

• a large fraction of the tasks currently performed by knowledge workers;
• an extreme concentration of financial resources;
• systems that underlie a majority of government functions and market decisions;
• autonomous weapons deployed in military and police contexts;
• mass surveillance over the physical and digital worlds;
• the information environment and mass culture;
• a large fraction of cutting-edge scientific research and development.

The above sources of power could become extremely concentrated in a small group of people through a few mechanisms:

• Coalition substitution and institutional capture: historically, those seeking centralized power needed the cooperation of employees, soldiers, bureaucrats, judges, intelligence officers, and so on. If these roles become mostly automated, we’ll lose much of the protective mechanism of individual humans being able to object, and it will be easier to expand unilateral control.
• Exclusive access: those in command of AI systems could use their position and resources to secure exclusive access to the most advanced AI capabilities, and entrench control over the most consequential deployments. Advanced models could surpass human experts in military R&D, cyber-offense, political and legal strategy, persuasion, and AI development itself. Consequential deployments include the military, domestic surveillance, government bureaucracy, legal systems, decision-making at large corporations, and global markets.
• Models with singular loyalties: a singularly loyal model pursues the interests of a single person or small group (the ‘principal’) above any other allegiances, moral considerations, or legal constraints (except when abiding by them is instrumentally useful to the principal). These loyalties may be overt or secret. Secret loyalties introduce novel threats, such as a model becoming widely deployed across governments and markets while introducing millions of subtle steers that strengthen the principal and weaken their adversaries.

We lack detailed forecasts of how power concentration could unfold. But we think it’s plausible that hard-to-reverse steps in these threat models could occur within the next five years. Capability trends on autonomous agentic tasks, the possibility of automating AI R&D itself, and broader uncertainty suggest near-term progress on power-relevant capabilities is plausible.

If power became extremely concentrated, it could translate into a global regime that could last for many generations into the future. Sufficiently advanced AI could encode a complex and nuanced set of goals and views that become entrenched by overwhelming economic and military dominance. Authoritarian institutions can end when leaders age and die, foreign powers intervene, or citizens rebel. A global regime formed through AI-enabled power concentration could become resistant to these sources of change.

In sum, we think it is plausible that:

• 1. Transformative AI could enable a small group of people to gain extreme control over the lives of everybody else.
• 2. Key developments in the establishment of this control could occur within the next five years.
• 3. If we enter such a world, it may be extremely difficult to break out of it.

Extreme power concentration may be one of the most important and neglected problems of our time. We want to fund projects and people to prevent it from happening.

---

Request for Proposals

There are two tracks: Grants and Career Funding.

• Grants will fund projects, programs, new teams, and new organizations.
• Career Funding will fund people to develop or transition their skills to work on risks from power concentration. This could include course buy-outs, post-docs, sabbaticals, training, independent study, or transition stipends.

---

Grants

This problem sprawls across disciplines, and our current understanding will likely prove incomplete and partly wrong. We want to be open-minded about the best approaches, and support work across a wide portfolio.

However, we expect excellent proposals will fulfill the characteristics below. We also have priority areas which we think deserve special attention.

Note: Proposals that are partisan or aimed at influencing elections are not eligible for this RFP.

Characteristics of excellent proposals

• Targeted.
  • Excellent proposals focus tightly on the specific problem of extreme power concentration enabled by AI, rather than adjacent issues like AI safety or existing inequalities, or technology policy writ large.
  • We are looking for work that addresses a specific mechanism through which AI could lead to an extreme concentration of power, or a defense against this happening. Such defenses do not need to be AI-specific, since many checks and balances apply generally (though near-term AI may place these measures under extraordinary pressure).
  • Proposals should articulate a clear theory of change—a defensible story for how the work will help understand or reduce the risks above. We recognize how hard it is to form crisp, confident theories of change in this space. We regularly support projects that may not pan out, and we expect many grantees to pivot as they learn. But we want applicants to think hard about their theory of change when choosing and pursuing their proposal.
• Rapid.
  • The AI landscape changes quickly, and important developments could occur on short timelines.
  • Excellent proposals have short feedback loops, produce useful outputs in months rather than years, and can adapt as the situation evolves. But as noted earlier, we’re open to supporting projects for long durations and at scale.
• Ambitious.
  • The scale of the problem is vast. Excellent proposals have the ambition to face up to that challenge and make real progress on tractable sub-problems.
  • They should not merely aim to marginally advance the literature or have meetings.
  • Longview has resources to back ambitious work if the case is justified. We encourage you to aim high.
• Forward-looking.
  • The world has been surprised by the speed of AI progress, and still struggles to keep pace.
  • Proposals should not fixate on AI at the capability level of mid-2026, despite that level being sufficient to cause novel problems in many fields.
  • Instead, proposals should, at minimum, consider the ongoing trend in capabilities and the implications of those trends continuing through the rest of the decade. Much more rapid progress is plausible (1, 2, 3), including the possibility of reaching AI systems that are superior to the best human experts across nearly all cognitive tasks.
  • Most areas listed below already have active research communities; this RFP seeks to push them forward by taking the radical implications of continued AI progress seriously.

Priority Areas

We are particularly excited to fund work in the twelve areas below. Proposals outside these areas will still be considered, but we are less likely to fund them.

The areas aren’t in priority order. If you have ideas in multiple areas, you should prioritize the proposals you are most excited about and feel most suited to. You can also submit multiple proposals—if you would not be able to do all of them you can just flag that in your applications.

The listed examples are just illustrations. We hope the RFP will draw out a diverse range of disciplines to surface the many promising ideas we haven’t thought of.

(1) Foundational research

We still have only a nascent understanding of the threat and the landscape of solutions. There are probably novel angles and ideas for the field that could be found through more researcher time.

Proposals in this area could include threat modeling, prioritization of interventions, and novel institutional designs (such as public-private structures that keep control distributed, or proposals for regulating AI without concentrating power). It could also include fleshing out the details of the problem or articulating positive visions to aim for, such as via legal analysis, institutional economics, political philosophy, or other fields.

(2) Situation monitoring

Important developments are likely to be kept secret, or may be hard to pick out from noise. We need defenders to be able to react quickly and appropriately, and we need the public to be well informed to stand as a check against powerful entities. Developments to track may include significant changes to:

• the formal or informal relationships between the government and frontier developers, especially when pressure is applied or agreements are intended to be kept secret;
• how AI is actually being used in critical government functions, and the regulations or internal policies guiding such use;
• relevant changes to the internal policies or use patterns of frontier companies.

Proposals in this area could include setting up independent analytical, OSINT capacity, or legal analysis.

(3) AI integrity and secret loyalties

Frontier AI systems could be compromised to advance an actor’s interests — an adversarial state, say, or a small group inside an AI company — without those making or using the model noticing.

Proposals in this area could include developing methods of detecting secret loyalties and security measures that could make attempts much more difficult, such as:

• hashing model weights and verifying them against an independent registry,
• auditable logs of training data and processes,
• data access controls,
• logging system interactions and outputs,
• weight access controls,
• compute use transparency,
• red-teaming
• and more.

Such mechanisms could also be misused as a tool for state control over model training, so applicants must carefully consider dual-use risks.

(4) Frontier company policies against extreme power concentration

The actions of frontier companies are central to this problem. Either regulation or the companies’ own policies will determine those actions. We think there is valuable work in designing voluntary policies, researching regulatory options, and engaging in educational communication with policymakers and the public to ensure frontier companies take the most responsible paths. Such decisions include:

• figuring out the details of prospective ‘red-lines‘ on what systems they provide to governments, under what conditions, and according to what publicly visible triggers;
• developing and publicly disclosing model specifications for models intended for government, military, or internal use;
• installing and auditing the measures for AI integrity mentioned in area 3 above;
• among many others.

(5) Law-following AI

Making AI agents follow the law seems like a minimum defense against mass misuse in the economy or government, and there is much to do to progress towards it. But the full intervention isn’t a clear-cut technical problem. How should AI operationalize legality? An AI may be given an order with apparent justification, but which the model believes a court would declare unlawful if it were reviewed. Should the model refuse a facially valid order, or assume it can predict later judicial opinion? A model might face a domestically lawful order that would likely cause mass harm, enable later law-breaking, break international law, or concentrate power. How should it balance these trade-offs?

Proposals in this area could include:

• tackling the open questions in this agenda;
• exploring and providing guidance on hard cases where law-following conflicts with other desiderata;
• making technical progress on implementing and ensuring AIs robustly follow the law, especially in high-stakes contexts;
• designing systems through which government-deployed AIs could whistleblow on illegal requests.

(6) Oversight of AI in government, including national security

Government AI use will likely carry the highest stakes. Governments tend to face limits in what they can do because human bureaucracies are slow and imperfect, and individuals can object, resign, or whistleblow. A government whose functions AI agents carry out may lose these limits, and if other branches don’t keep pace, checks and balances will weaken. We focus on democratic countries because affecting authoritarian governments’ use of AI seems much less tractable. However, we are open to proposals on the latter.

Proposals in this area could include:

• researching regulatory options on procurement frameworks, and engaging in educational communication with policymakers and the public;
• progress towards designating chat logs and agent traces as records with retention requirements (including purpose limitation and judicial access controls to minimize misuse);
• new legal, technical, and institutional tools for legislative and judicial oversight (including large-scale AI itself) to keep in the loop, especially for use of AI in defense and intelligence services;
• developing privacy-preserving auditing systems (including structured transparency);
• research and educational programs that support the capacity of watchdog elements such as inspectors general and legislative oversight committees;
• tackling open questions in this agenda.

(7) Preventing misuse of autonomous weapons

Remotely operated drones have become an essential part of modern warfare, and loitering munitions with autonomous targeting are currently deployed. Militaries are pushing for autonomy in engagement decisions, mission planning, force-level coordination, and strategic decision support, and an intelligence explosion could bring advances much faster than expected. Existing military doctrine is being rapidly rewritten for a world in which autonomous systems compose a large and growing fraction of fighting forces. If militaries treat autonomous weapons like any other system, and AI replaces more and more decision-makers, an ever-shrinking number of humans will hold control. This would create new single points of failure — vulnerable to adversaries disabling or taking them over, and to misuse, whether by individuals with privileged access or by external compromise.

We are most interested in this power concentration angle raised by future systems with significant advances in autonomy, rather than current issues of meaningful human control, international humanitarian law compliance, non-state proliferation, crisis escalation, and so on.

Proposals in this area could include figuring out appropriate chains of command for networks of autonomous weapons (e.g., avoiding overly centralized control), safeguards against misuse, norm-building, and development and advocacy for international agreements.

(8) Protecting against AI-enabled mass surveillance

Governments and businesses now collect data in bulk across the physical and digital worlds—faster than they can analyze it. Powerful AI could dramatically reduce the cost of targeted surveillance, enabling intrusive monitoring of any and all political opponents, from protestors to judges. Near-ubiquitous surveillance may be achievable today in advanced economies, even before hardware increases, and especially if end-to-end encryption is dissolved. Rights of privacy, free expression, and free assembly could suddenly face an extreme threat, for which we are unprepared.

Proposals in this area could include:

• developing model statutory frameworks and analysis on AI-enabled domestic surveillance (e.g. 1, 2);
• transparency requirements for effective oversight (such as disclosure of government purchases from commercial brokers) and/or using AI to enhance legislative, judicial, and public oversight mechanisms;
• exploring privacy-preserving alternatives to warrantless mass surveillance;
• frameworks that gain the security benefits of AI surveillance without enabling abuse.

(9) Political economy for a changed world

Transformative AI is likely to present dramatic challenges to our ordinary economics and politics. AI could accelerate the long-running decline in labor’s share of income, depress wages for workers whose tasks are most substitutable, and may produce mass unemployment. Government revenue could depend on AI income and become fiscally disconnected from its citizens (an intelligence resource curse). If society can navigate these and other challenges wisely, it will be better placed to resist extreme concentrations of power. If it cannot, despotism becomes more likely.

Proposals in this area could include modeling the above threats to stability, empirical data collection and analysis, and institutional designs that provide more robust societal systems. In particular, sudden and dramatic shifts could cause countries to nationalize AI development, presenting significant risks of power concentration. What alternative mechanisms or designs could provide security without these risks? We may need to analyse, design and debate novel market mechanisms or taxation and redistribution systems. Civil society and our political discourse need to be as informed as possible about the scenarios that could quickly arrive. Examples of relevant work can be found here: 1, 2, 3, 4.

(10) AI tools to inform and empower citizens

AI could saturate and distort the information environment. But AI could also uplift our ability to monitor, understand, and make decisions about the world around us. AI tools could improve political discourse if they were well-calibrated, properly uncertain, accurate about the positions they represent, faithful to their user, and supportive of collective deliberation and compromise. AI may also serve or undermine other public dynamics; how can it be steered towards the former? Even without mass adoption, AI tools could supply a balanced, expert-level research team for every journalist, judge, legislative staffer, civil society leader, and advocate. By default, such tools may be undersupplied by the market, and not a design priority of frontier developers.

Proposals in this area could include evaluations and benchmarks for frontier models that focus on epistemic virtues, tools and platforms for collective deliberation, or development of faithful AI delegates.

(11) Building compute in democracies

Power is less likely to become dangerously concentrated if the compute underpinning advanced AI is distributed across allied democracies rather than held by a single country, and if that group’s compute and access to capabilities stay well ahead of authoritarian rivals. Both aims are served by ambitious compute build-outs in allied democracies rather than in authoritarian states.

Proposals in this area could include research and analysis on the geographic distribution of compute and its implications for extreme power concentration.

(12) Field building

This area is nascent, and not enough people work on it directly. Growing this field is a high-leverage intervention, and we’d be excited to back projects that help people learn about it and start contributing. Proposals could include:

• training and mentorship programs;
• events (such as conferences, seminars, workshops, hackathons);
• public resources and content (like blogs, books, podcasts, videos, and so on).

Assessment criteria

We aim to assess proposals based on the following:

• Theory of change. Does the proposal have a plausible theory of change? That is, a credible account of how, if successful, the work would better understand or reduce the likelihood or severity of extreme power concentration. Will the funding enable the right actions to be taken, and will those actions accomplish the goal?
• Track record. Does the applicant have evidence of succeeding at comparable work before? Or evidence of future promise for the area? This problem is nascent, so we will consider a wide range of past achievements and present indicators to be relevant.
• Strategic judgment. Power concentration is a fast-moving, uncertain, and sensitive area. We look for applicants who can make sound decisions under those conditions, such as showing reasoning transparency, scope sensitivity, and adaptability to an AI landscape that is transforming every few months.
• Project risks. Does the application identify the most significant failure modes and downside risks of the proposal? Work in this area could be misinterpreted and politicized, or could backfire in other ways. We don’t require zero-risk projects, but we expect applicants to show that they understand the risks and how to mitigate them. For example, proposals should consider how their outputs could be misused and whether they could worsen other kinds of extreme risk (such as risks from power-seeking AI systems).
• Cost-effectiveness. Our resources can do the most good if the cost of projects is commensurate with their expected impact. This doesn’t mean budgets need to be shoestring; it means that high budgets have to be justified by high expected impact. But feel free to submit ambitious proposals — including both a mainline and an ambitious version of a project with corresponding budgets. We are usually not at all concerned by what percentage of the budget is operating expenses; the right level is whatever’s optimal for achieving impact.

Application process and deadline

To apply, please fill out the application here. The application requires the following:

• Basic grant details and logistics questions (e.g., biographical details for those involved).
• Resumes, CVs, or LinkedIn pages for key personnel that include prior employment, educational background, key achievements, and references.
• Project proposal (300 words).
• Mainline budget amount.

For questions about the application, please email power-rfp@longview.org. The application deadline is July 2, 2026.

---

Career Funding

Kinds of support

We want to grow the field of people working on this problem, and funding projects, programs, and organizations is a key way to do that. But not everyone will have a crisp proposal they want to commit to, and some may want to explore and skill up in the area before making the jump. We are keen to offer Career Funding to support individuals planning to transition to high-impact work in the area.

The Career Funding can consist of a stipend and/or expenses. The stipend will be sized by the length of time, cost of living, and how promising the applicant’s prior experience and plans seem for this space. A wide range of expenses can be supported, including expenses for travel, tickets to conferences, contractors, and useful subscriptions, books, or courses.

Profiles we’d be excited about

• A founder-type who wants to skill up in the space to prepare for launching a new organization in the area.
• A talented generalist who wants to switch to working on power concentration full-time, including building and running new teams and organizations.
• A post-doc seeking financial support to study an aspect of AI-enabled concentration of power (for example in law, political science, computer science, or economics).
• A faculty member in law, political science, economics, computer science, or a related field seeking a course buy-out or sabbatical support to redirect research time toward an aspect of AI-enabled concentration of power.
• A technical AI researcher who would like to explore technical work on law-following AI, AI integrity, or oversight tooling in order to build a full-time role.
• A legal scholar or practitioner seeking to develop draft model legislation, doctrinal frameworks, or litigation theory related to AI oversight, federal AI deployment, or surveillance.
• A former national security professional, military officer, or intelligence community veteran interested in applying their expertise to oversight of military and intelligence AI, autonomous weapons doctrine, or legislative oversight of national security AI.
• A former civil servant, legislative staffer, or policy researcher seeking to develop concrete reforms for government AI procurement, reporting, or inter-branch oversight mechanisms.
• An economist or political economist interested in exploring AI’s implications for labor markets, taxation, redistribution, or resource-curse dynamics for intelligence.
• An investigative journalist, OSINT researcher, or communications professional pivoting to work on monitoring AI labs and government AI use, or on raising public awareness of AI-related risks to civil liberties.

Selection criteria

Applications will be selected primarily on the basis of the following criteria:

• Interest in AI-enabled concentration of power and related issues. Ideal applicants demonstrate they understand how AI interacts with extreme power concentration, and commit to high-impact work with plausible practical relevance.
• Evidence of potential for impact, as demonstrated by prior achievements, employment record, academic achievement, and references.
• Promising planned activities and potential future work activities.

Applications must be compliant with all relevant local laws, and applicants must be located in countries where Longview is able to make grants, as listed in the application form.

Application process and deadline

To apply, please fill out the application here. The application requires the following:

• Basic biographical information and logistics questions.
• Resume, CV, or LinkedIn page that includes prior employment, educational background, and key achievements.
• Discussion of what motivates you to want to work on this topic (200 words).
• Description of how you will spend your time during the Career Funding (200 words).
• Discussion of what you might do after the Career Funding (100 words).
• Funding requested and timeline.

For questions about the application, please email power-rfp@longview.org. The application deadline is July 2, 2026.

---

Feedback

We intend to evaluate proposals on a rolling basis. We are unlikely to provide feedback in most instances, as we want to focus capacity on developing the highest-impact proposals as quickly as we can.

---

Eligibility (for both tracks)

Proposals that are partisan or aimed at influencing elections are not eligible for this RFP.

Longview is unable to select individuals or entities operating in the following jurisdictions: Afghanistan, Belarus, Bosnia and Herzegovina, Burma/Myanmar, Burundi, Central African Republic, Chad, Congo, Cuba, Democratic People’s Republic of Korea (North Korea), Democratic Republic of the Congo, Eritrea, Ethiopia, Guinea, Guinea-Bissau, Haiti, Hong Kong, Iran, Iraq, Lebanon, Libya, Mali, Myanmar, Nicaragua, Russia, Somalia, South Sudan, Sudan (including Darfur), Syria, Ukraine — restricted regions only (Crimea, Donetsk, Luhansk, Kherson, Zaporizhzhia), Venezuela, Yemen, Zimbabwe.

Longview is unable to grant to individuals in India or China. Longview may grant to Indian non-governmental organizations registered under the Foreign Contribution (Regulation) Act.

Longview reserves the right to exclude additional jurisdictions at any time based on regulatory requirements, changes in law or policy, or other legal or compliance considerations, without prior notice.

Award of funding is contingent on satisfactory completion of Longview’s due diligence and the grantee’s execution of and compliance with Longview’s grant terms.