A public map of who shapes U.S. AI policy, what they believe, and how they are connected. We’ve mapped 2,041 people and organizations and 2,978 relationships, covering affiliations, funding, policy positions, AI timelines, risk views, and sources.
Every policy intervention implicitly assumes a threat model. We want to systematically map these threat models–their assumptions, causal pathways, and the measurement systems needed to track them.
We plan to interview losing primary candidates about their theory of change and diagnosis of failure points to inform strategy for the 2028 presidential election. We suspect losing candidates will be more candid about their experience, especially about lobbying interests that approached their campaigns.
We want to build strategic relationships with pro-democracy groups, corporate and constitutional law scholars, and other fields that are motivated to prevent concentration of power. By illustrating the risks of AI-enabled power concentration through pathways like democratic backsliding, coups, and capital accumulation, we hope to increase support for robust technical governance and policy interventions.
A pilot proposal for structured transparency between frontier labs and government statistical agencies, to enable real-time data sharing about AI’s impacts on the economy.
Today’s corporate structures don’t hold AI megacorps accountable for their effects on the public. We’re exploring new institutional forms, such as an enforceable public fiduciary, that bind firms to account for the public interest through both internal governance and verifiable external checks.
Different institutional forms are suited to different problem spaces (see: FROs for building shared scientific research infrastructure; standards bodies for decentralized coordination). Applied to AI, this points to concrete gaps–for instance, an IAEA-style scientific body for verifying compute and training claims, a NTSB-style bureau for investigating AI incidents across companies, and a FRO-style entity for funding shared safety infrastructure like evals and interpretability tools. This project would build a working catalog of problems matched to institutional types and propose new forms where necessary.
