OpenAI’s AI economic-impact grants will fund 14 projects across five markets

The projects receiving support

Biosecurity information-sharing

The Nuclear Threat Initiative will assess whether secure, cross-border sharing of information about AI capabilities, risks, and mitigations is legally feasible. It will then develop a basic international information-sharing architecture focused on AI and biosecurity risks.

Coordination around recursively self-improving AI

The Institute for Security and Technology will create a framework for measuring and coordinating responses to uncontrolled recursively self-improving AI. Its work will cover definitions and indicators of self-modification, cross-lab incident taxonomies, and escalation criteria and pathways for coordination between industry and government.

Preparing for workforce disruption

A bipartisan commission involving the American Enterprise Institute and the Urban Institute will develop low-, medium-, and high-disruption scenarios for AI’s effects on jobs and skills. The project will connect observable indicators of disruption to policy playbooks that governments could use as conditions change.

Portable benefits and “livelihood insurance”

The Progressive Policy Institute will design portable, person-based benefits covering areas such as retirement, health, leave, education, training, and disability. It will also prototype “livelihood insurance” that could be triggered by occupation-wide change, while identifying the legislative and technical steps needed to implement these ideas.

How AI could change public revenue

The Tax Foundation will model how AI adoption could shift the sources of public revenue among labor income, corporate profits, capital gains, and other categories. Its planned white paper will assess tax options against criteria including neutrality, simplicity, transparency, stability, efficiency, innovation, competitiveness, and fiscal resilience.

Who receives AI-driven productivity gains in Europe

The Centre for European Policy Studies will study how AI-related productivity gains are distributed among workers, firms, sectors, regions, and countries. It will benchmark the European Union against the United States and develop policy principles for a renewed European social-investment model and adaptive safety nets. Planned outputs include a study, visualizations, and a stakeholder workshop.

Privacy-preserving simulations for public policy

Nanyang Technological University will build and evaluate privacy-preserving large language model agents that use individual financial-transaction data to derive behavioral profiles. The aim is to simulate how households might respond to government transfers and industrial policies before those policies are deployed.

The project’s planned deliverables include an auditable prototype, a privacy-preserving profile schema, prediction reports, and guardrails for use.

AI infrastructure for Brazil’s public health system

Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo will conduct a controlled, pre-deployment evaluation of an AI clinical-information infrastructure prototype for Brazil’s public Unified Health System, known as SUS.

Using synthetic or appropriately de-identified data, the evaluation will examine whether the system can reduce fragmented record review, improve clinical workflow and evidence-based risk stratification, and help patients with navigation, prescription renewals, and symptom escalation. The project is designed to retain clinician judgment and human oversight.

Why the announcement matters

The program is notable less for a single product launch than for the range of policy and institutional questions it puts into funded research. The projects address both the potential benefits of AI—such as improved public services and better policy modeling—and the risks associated with labor disruption, fiscal pressure, biosecurity, privacy, and advanced AI safety.

OpenAI describes the grants as a starting point for building evidence and encouraging debate about the policies needed for the “Intelligence Age.” Because the work is scheduled to run for six months, the expected 2027 reporting will be important for distinguishing initial proposals and prototypes from approaches that have demonstrated practical value.