Written and maintained by CASRAI Editorial Board
Last updated
What AIPI is
The AI Policy Institute (AIPI) — formally the Artificial Intelligence Policy Institute — is a US nonprofit that researches and polls public opinion on advanced AI systems and publishes the results for policymakers, journalists and the public. On its own site it describes itself as “an independent research organization dedicated to advancing public understanding of the trajectory, capabilities, and implications of the most powerful current and future AI systems.”
AIPI states it is a “donor-supported 501(c)(3) non-profit.” It is not a government body, an AI evaluation lab, or an industry association, and it has no statutory role in AI regulation. Its output is polling data and policy analysis, not binding standards or technical model assessments.
Who runs it
AIPI was founded by Daniel Colson, who is its executive director. AIPI’s public pages do not disclose a founding date or a full staff/board listing beyond a “Team” section referenced from its About page, so those details are not stated here.
What AIPI does
- Public opinion tracking. AIPI commissions surveys, working with polling firms, that ask US voters about AI development, specific risks, and specific policy proposals — for example support for testing requirements, liability rules, or siting conditions on AI infrastructure.
- Policy research. It publishes what it describes as “comprehensive reports and policy proposals” analysing AI governance questions.
- Media and policymaker engagement. It positions its polling and analysis as a resource for government officials and journalists covering AI policy.
A representative example
A useful illustration of AIPI’s output is its poll Opposition to Local Data Centers Falls When AI Safety Guardrails are Attached, published 4 August 2026 and fielded 10–11 June 2026 among 1,007 likely US voters (margin of error 4.2 points). It found that opposition to a hypothetical local data center fell from 57% to 38% once AI safety guardrails were attached to the proposal, that majorities of both Democrats (72%) and Republicans (77%) would accept a guardrailed data center in their community, and that when voters were asked to choose directly between a regulated and an unregulated facility, they preferred the regulated one 74% to 6%. The finding is typical of AIPI’s work: a specific, fielded poll used to argue that safety conditions are not a political liability for AI infrastructure projects.
How it differs from other organisations in this cluster
AIPI is a public-opinion and policy-research organisation, not a technical evaluator or a standards body. That distinguishes it from other institutions covered in this cluster: METR runs technical evaluations of frontier models’ autonomous capabilities; the Frontier Model Forum is an industry body that coordinates safety commitments among AI developers; and the International Network of AI Safety Institutes connects government safety institutes. AIPI’s role in that landscape is to measure and publish what the voting public thinks about AI risk and AI regulation, and to use that data to argue for particular policy positions.
A note on the domain
AIPI’s site is theaipi.org. A plausible-looking alternative, aipolicyinstitute.org, is not affiliated with the organisation; as of September 2026 it resolves to a parked, for-sale domain page and is not a source for information about AIPI.
How CASRAI’s NIKOLAI tracks this
The central variable in AIPI’s data-center poll above — “AI safety guardrails” — is the same concept CASRAI catalogs as a safeguard in NIKOLAI, CASRAI’s own independent dictionary of frontier-AI-safety terminology, under its Mitigations and security track: a technical or procedural measure meant to reduce misuse or misalignment risk, identified by type, target risk, and deployment scope. AIPI measures public appetite for guardrails in the abstract; NIKOLAI tracks how individual developers and deployers actually define and scope them in practice — as unendorsed, shadow mappings unless an organization files its own Mapping Declaration confirming its usage. AIPI itself has no NIKOLAI declaration, since it is a polling and policy-research organization rather than an AI developer or deployer.
