Miles Brundage is an American AI policy researcher. He joined OpenAI in 2018 and eventually headed its Policy Research team and then its AGI Readiness function, before departing in October 2024 and becoming an independent researcher and adviser. He is the lead author of "The Malicious Use of Artificial Intelligence" (2018) and a co-author of "Toward Trustworthy AI Development" (2020).
Type: Individual (policy researcher) Affiliations: Independent (2024–present); formerly OpenAI (Senior Advisor for AGI Readiness; head of Policy Research); formerly Future of Humanity Institute, Oxford
Background and roles
Brundage joined OpenAI in 2018. Over the following six years he headed the Policy Research team and then the AGI Readiness function, and was one of the central continuity figures on OpenAI's externally facing AI governance posture, responsible for much of what was publicly visible in OpenAI's approach to evaluations, publishing norms, and preparedness. At OpenAI he was involved in early publication-norm debates, including the staged release of GPT-2, the earliest iterations of the Preparedness Framework, and external engagement with institutions in the AI-safety-institute network. He had earlier worked at the Future of Humanity Institute at Oxford.
He left OpenAI in October 2024 with a public statement that he no longer felt able to do the research he considered most important from inside the organisation. He is often cited as the person who carried most of the institutional memory of OpenAI's earlier, more safety-forward policy culture, and his departure has been treated as a marker for that culture's contraction. His departure came amid a cohort of 2024 OpenAI departures, alongside Leopold Aschenbrenner, Jan Leike, and Daniel Kokotajlo, that reshaped public perception of the lab's safety posture.
Since 2024 he has worked as an independent researcher and adviser on AI policy, active on publication, commentary, and government and regulator engagement outside any lab affiliation, including post-departure public writing on what a responsible "AGI readiness" agenda looks like, gaps in the OpenAI approach, what external regulation should require, and congressional engagement.
Notable publications
Brundage is the lead author of "The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation" (Brundage et al., 2018), an early mapping of AI misuse risks. He is a co-author of "Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims" (Brundage et al., 2020), a paper on AI auditing, red-teaming, incident sharing, and bias bounties. He is also among the authors of "Release Strategies and the Social Impacts of Language Models" (2019), associated with the GPT-2 release.
Positions and statements
In his October 2024 departure statement, Brundage argued that "no one is ready for AGI, including OpenAI," framing neither labs nor governments as prepared for the capability trajectory he views as plausible.
He advocates for substantial public-sector capacity, including AI safety institutes, NIST, and Congressional staff, on the view that self-regulation is insufficient. He is a long-standing advocate of auditing, verifiable claims, and incident-sharing, mechanisms he co-authored in the 2020 "Toward Trustworthy AI Development" paper and which he describes as informing AI-safety-institute test-access arrangements.
On existential risk, Brundage holds a middle-ground position skeptical of both doomer and dismissive framings: he treats existential risk as plausible and worth governing against, while arguing that current discourse over-indexes on vivid scenarios at the expense of tractable policy.
Relationships
- related: OpenAI — former employer; departure is a policy-culture data point
- related: Leopold Aschenbrenner, Jan Leike, Daniel Kokotajlo — the cohort of 2024 OpenAI departures that reshaped public perception of the lab's safety posture
- supports: AI Safety Cases and Frameworks, Compute Governance
- related: US AI Safety Institute (NIST AISI), UK AI Safety Institute (AI Security Institute)