OpenAI Funds 14 Independent Projects to Stress-Test Its AI Policy Agenda

OpenAI is awarding grants to 14 projects led by independent organizations across the United States, the European Union, Brazil, Singapore, and South Korea, committing a total of $1 million in funding and up to $1 million in API credits. The company frames the program as a follow-through on a pledge it made in April 2026, when it published its policy paper "Industrial Policy for the Intelligence Age": to support outside groups in testing, challenging, and building on the ideas OpenAI put forward.

The stakes are straightforward. OpenAI says ChatGPT now has 1 billion users worldwide, spanning geographies, income levels, and demographic groups, and claims it has more adult users both under and over the age of 30 than any other frontier lab. But in its own announcement, the company concedes that broad access to the technology is only part of the equation. "Whether AI ultimately serves the many rather than the few will also depend on the choices societies make about how to deploy it and distribute its benefits," OpenAI writes.

The company's argument is that the policy response needs to be "as ambitious as the technological change itself"—which means new approaches rather than adjustments to existing policy. It also explicitly steps back from claiming sole authorship of those choices. "These choices should not be made by technology companies alone," OpenAI states. "In democratic societies, they should be shaped through democratic institutions and public debate."

The grant program is OpenAI's mechanism for widening that debate. Independent research and policy organizations, the company argues, can develop and test new approaches, challenge assumptions—including OpenAI's—and act as "laboratories for ideas that governments can adapt to their own communities and circumstances." More than 400 people and organizations responded to OpenAI's call for proposals. Each applicant identified an idea from the April paper and pitched a project that could contribute to public understanding and policy development.

Two questions, fourteen answers

The selected projects cluster around two broad questions: how AI can broaden economic opportunity, and how societies can build resilience as capabilities advance. Some will produce research and policy models; others will build prototypes, datasets, and frameworks intended for real-world testing. All projects run for six months, with results reported in 2027.

The economic opportunity portfolio

The American Enterprise Institute will run the AEI-Urban Bipartisan Commission on Artificial Intelligence and the Future of the American Workforce, developing low-, moderate-, and high-disruption scenarios for AI's effects on employment and skills. The commission will connect observable indicators of disruption to concrete policy playbooks.

The Centre for European Policy Studies takes on the distribution question directly. Its "Sharing the AI Dividend" project will examine how AI-driven productivity gains are distributed across workers, firms, sectors, regions, and countries in Europe, benchmarking the EU against the United States. Building on a synthesis of existing empirical evidence, CEPS will develop policy principles for a renewed European social investment model and adaptive safety nets. Deliverables include a policy study, visualizations, and a stakeholder validation workshop.

The European Centre for International Political Economy will develop a practical framework for a "Right to AI." Its "From Labour to Productive Ownership" project will compare models including employee ownership, citizen investment, pension-based ownership, social wealth funds, and intellectual-property participation, producing a final report, shorter insights, and a public webinar.

The Abundance Institute addresses the energy dimension: its "Energy Abundance, State by State" project will compare how US states can expand generation and transmission in response to data center demand while producing measurable local benefits. Outputs include a state policy framework, historical analysis, and a public Data Center Atlas policy layer.

The Progressive Policy Institute will design a person-based benefits system covering retirement, health, leave, education, training, and disability across different forms of work. Its most concrete deliverable is a prototype of "livelihood insurance" triggered by broad changes in an occupation, along with an identification of the legislative and technical changes required to implement the model.

The Tax Foundation will analyze how AI adoption could shift the balance among labor income, corporate profits, capital gains, and other sources of public revenue. A final white paper with quantitative analysis will assess taxation options against neutrality, simplicity, transparency, stability, and economic efficiency, explicitly considering innovation, adoption, competition, competitiveness, and fiscal resilience.

The Windfall Trust will scale a network of working groups in the United States, United Kingdom, Canada, the European Union, and Latin America to examine fiscal, labor-market, welfare, and competitiveness questions under different AI scenarios. It will also launch an International Working Group focused entirely on international economic preparedness and coordination, connecting economists, policy experts, pollsters, and regional partners, and will publish public syntheses of the main working groups' takeaways.

Research capacity and public health

Brazil's Instituto de Matemática Pura e Aplicada will study what research institutions need to turn AI access into meaningful scientific progress—including engineering support, training, governance, compute, and local infrastructure. Working with mathematicians in Brazil, IMPA will develop a transferable measurement and cost framework, practical guidance for trustworthy AI use in mathematics, and a public report evaluating the Right to AI and distributed scientific discovery.

Also in Brazil, the Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo will conduct a controlled, pre-deployment evaluation of an AI-enabled clinical information infrastructure prototype designed for Brazil's Unified Health System, SUS. Using only synthetic or appropriately de-identified data, the project will assess whether people-first clinical AI can reduce fragmented record review, improve clinician workflows, support evidence-based care and risk stratification, help patients navigate health services, manage prescription-renewal workflows, and identify warning symptoms requiring escalation—all while, in OpenAI's words, "preserving physician judgment and human oversight."

Frontier risk and governance

The Institute for Security and Technology will build an operational framework for uncontrolled recursively self-improving (RSI) AI systems. The project will develop technically grounded definitions, observable indicators of uncontrolled self-modification, incident taxonomies for cross-lab communication, and escalation criteria that map technical signals to coordinated response decisions. Outputs include an incident-classification schema, escalation pathways for industry-government coordination, and a public executive summary.

The Nuclear Threat Initiative's bio team will test the legal feasibility of secure international sharing of AI capabilities, risks, and mitigations, then design a basic information-sharing architecture to enhance international AIxBio safety and security.

The Council on Strategic Risks' Converging Risks Lab will examine how national-security professionals can build institutional capacity in frontier-AI safety and governance. Through fellowship training and engagement with frontier AI labs and safety researchers, the project will produce public-facing analysis on oversight and preserving meaningful human control as AI capabilities advance.

Government simulation and legislative oversight

Nanyang Technological University in Singapore will develop and evaluate an AI-agent framework that transforms individual-level financial transaction data into privacy-preserving behavioral profiles to simulate how households would respond to government transfers and industrial policies before implementation. The simulations will be validated against observed responses to prior policies and established statistical benchmarks. Deliverables include a prototype, a privacy-preserving profile schema, auditable reports explaining the basis for each prediction and how changes in policy design would alter it, and guardrails for responsible government use.

Yonsei University will test whether transparent, human-validated AI methods can help assess the quality of oversight in South Korea's National Assembly. The initial phase will produce an analytical report, pilot dataset, and codebook examining evidence use, polarization, and the quality of legislative oversight.

Why it matters

The program is notable for what it signals about OpenAI's positioning in policy debates. In "Industrial Policy for the Intelligence Age," the company argued that governments should be prepared to use the full policy toolbox—from research funding and workforce development to market-shaping tools and targeted regulation—to manage a technological transition of this scale. Funding outside institutions to stress-test that argument costs OpenAI $1 million, a negligible sum against its operations, but it buys the company a distributed network of credible researchers examining questions it has itself framed.

The grantee list also spans a wide ideological and geographic range—AEI and the Progressive Policy Institute, US think tanks and Brazilian hospitals, Singaporean engineers and South Korean political scientists—which suggests an effort to make the "Intelligence Age" policy frame durable across political cycles and jurisdictions.

OpenAI itself keeps expectations modest. "These projects are a beginning," the company writes, "and we hope the results will inform a much broader public debate about the policies the Intelligence Age requires." Whether the results, due in 2027, challenge OpenAI's assumptions as readily as the company invites them to will be the real test of the program's independence.