Miles Brundage

wiki
, , , ,
1

AI policy researcher, former Head of Policy Research and Senior Advisor for AGI Readiness at OpenAI, and founding Executive Director of AVERI — the nonprofit institute built on the principle that AI companies should not be allowed to grade their own homework.

Profile

Field Detail
Nationality American
Current Institution AVERI — AI Verification and Evaluation Research Institute (Executive Director)
Other Affiliations Non-Resident Senior Fellow, Institute for Progress; Member, AI Governance Forum (CNAS); Member, AI Policy & Governance Working Group (IAS/Princeton); Advisor, Epoch AI; Advisor, RAND Corporation
Research Areas AI Auditing and Verification, Frontier AI Regulation, Compute Governance, Assessment of AI Progress, Economic Impacts of AI, AI Grand Strategy
PhD Human and Social Dimensions of Science and Technology, Arizona State University, 2019
PhD Committee David Guston (chair); Erik Fisher; Joanna Bryson; Lauren Keeler
Personal Website milesbrundage.com
Substack milesbrundage.substack.com
X / Twitter @Miles_Brundage

Overview

Miles Brundage is an American AI policy researcher who spent six years at OpenAI — first as a research scientist, then as Head of Policy Research, and finally as Senior Advisor for AGI Readiness — before leaving in October 2024 to work independently. During his OpenAI tenure he was a central figure in shaping the company’s deployment practices: his teams initiated OpenAI’s external red-teaming program, produced the first several OpenAI system cards, and published research on topics ranging from the societal implications of language models to compute governance and frontier AI regulation. Before OpenAI he was a Research Fellow at Oxford’s Future of Humanity Institute, where he co-led the landmark 2018 report on the malicious use of AI — one of the first comprehensive frameworks for thinking about AI security threats, covered by the BBC, New York Times, and Wired. In January 2026 he launched AVERI (the AI Verification and Evaluation Research Institute), a nonprofit anchored in the argument that independent external auditing of frontier AI systems must become universal, and that AI companies should not evaluate their own safety. He is widely regarded as one of the most influential practitioner-voices in frontier AI policy.

Early Life & Education

B.A., Political Science — George Washington University, 2010
Brundage studied political science at GWU, where he pursued internships related to energy policy. His early interest in governance of emerging technologies, combined with energy policy work, set the trajectory toward AI governance.

ARPA-E / US Department of Energy (c. 2010–2012)
After graduation, Brundage worked for approximately two years at the Advanced Research Projects Agency – Energy (ARPA-E), the US government’s high-risk energy research agency housed within the Department of Energy. This experience working at the intersection of government, technology, and policy instilled an understanding of how federal agencies can shape technology trajectories — later relevant to his interest in the AI Safety Institute and compute governance.

Institute for Human and Machine Cognition (IHMC) — Internship
Brundage also interned at IHMC, a cognitive systems research institute, deepening his exposure to human-machine interaction questions.

Ph.D., Human and Social Dimensions of Science and Technology — Arizona State University, 2019
Brundage pursued his doctorate in ASU’s interdisciplinary program in the social studies of science and technology, supervised by David Guston (chair), Erik Fisher, Joanna Bryson, and Lauren Keeler. His graduate research was supported by the National Science Foundation, the Bipartisan Policy Center, and the Future of Life Institute. The program’s emphasis on responsible innovation, anticipatory governance, and the co-production of science and society gave his subsequent AI policy work a distinctive theoretical grounding unusual in a field often populated by technical researchers who migrated into policy.

Career

Future of Humanity Institute, University of Oxford — Research Fellow (2016–2018)

While completing his dissertation (completed 2019, though the FHI position overlapped), Brundage was a Research Fellow at Oxford’s Future of Humanity Institute — at the time the world’s leading academic center for existential risk and AI safety research. His most significant output from this period was the landmark 2018 report on malicious AI use.

“The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation” (February 2018): Co-led by Brundage with Shahar Avin, the report assembled 26 researchers from FHI, Cambridge’s Centre for the Study of Existential Risk, OpenAI, the Electronic Frontier Foundation, and other institutions — including Dario Amodei, Jack Clark, Helen Toner, and Paul Scharre. It provided the first comprehensive survey of AI-enabled security threats across digital, physical, and political domains; identified deepfakes, automated cyberattacks, and AI-assisted disinformation as priority risks; and proposed governance and technical responses. Covered by BBC, New York Times, Wired, and the Washington Post, it established Brundage as the field’s leading voice on AI misuse and security threats and was submitted to legislative bodies in the UK, EU, and US as reference material in subsequent AI regulation discussions.

OpenAI — Research Scientist, Head of Policy Research, Senior Advisor for AGI Readiness (2018–2024)

Brundage joined OpenAI in August 2018 as a research scientist on the Policy team, one of the first dedicated policy researchers at a frontier AI laboratory. Over six years and three roles, his responsibilities expanded to include leading the Policy Research team (which produced OpenAI’s system cards and external-facing policy documents) and then the AGI Readiness team (which assessed both OpenAI’s internal readiness for increasingly capable systems and the broader world’s readiness to govern them).

External red teaming program: Brundage’s teams launched OpenAI’s external red teaming program — structured adversarial evaluation by outside experts before model release — establishing the practice as part of OpenAI’s release process and contributing to the emerging norm of pre-deployment evaluation across the industry.

OpenAI System Cards: The policy and safety team Brundage led produced the first several OpenAI system cards — structured documents describing model capabilities, limitations, and safety evaluations accompanying major releases (GPT-2, GPT-3, Codex, GPT-4, ChatGPT, and others). These documents helped define the genre of frontier AI transparency reporting and set a precedent that other labs subsequently adopted.

Compute governance research: Brundage co-authored one of the field’s foundational papers on compute governance (arXiv:2402.08797), articulating the argument that computing hardware — uniquely detectable, excludable, quantifiable, and produced through a concentrated supply chain — is the most tractable point of regulatory leverage in the AI ecosystem.

AGI Readiness and levels framework: In his final role, Brundage was responsible for advising OpenAI executives and the board on the company’s readiness to steward increasingly powerful AI capabilities and on the world’s readiness to govern them. His team worked on fleshing out OpenAI’s “levels of AI” framework — a structured alternative to binary AGI thinking.

Departure (October 2024): Brundage left OpenAI on October 23, 2024, publishing a detailed Substack post explaining his decision. His stated reasons included the desire for greater independence and freedom to publish without industry review, concern about the perception of bias when working inside a frontier lab, and a view that industry-independent voices in AI policy were critically underrepresented. He was explicit that his departure was not driven by a negative assessment of OpenAI specifically but by a structural argument about where policy research is most impactful. He noted: “neither OpenAI nor any other frontier lab is ready, and the world is also not ready” for AGI — while clarifying this was not a controversial view within OpenAI’s leadership.

March 2025 public critique: After leaving, Brundage publicly responded to an OpenAI blog post praising it overall but accusing it of “rewriting the history” of the company’s deployment approach in a way that placed the burden of proof on safety concerns — a pointed criticism from someone with direct institutional memory of the policies in question.

Axon AI and Policing Technology Ethics Board (2018–2022)

Brundage served for four years as a member of Axon’s ethics board for AI and policing technology — one of the first corporate AI ethics boards for a company whose products directly intersect law enforcement. In June 2022, the board resigned en masse in protest over Axon’s proposal to develop a Taser-armed drone for school security — a case that became a prominent example of the limits of corporate ethics boards without structural enforcement power.

AVERI — Executive Director (January 2026–present)

In January 2026, Brundage formally launched the AI Verification and Evaluation Research Institute (AVERI), a nonprofit whose mission is to make third-party auditing of frontier AI systems effective and universal. The launch was covered exclusively by Fortune and coincided with the publication of a research paper co-authored by Brundage and more than thirty AI safety researchers and governance experts, laying out a detailed framework for how independent audits could work in practice.

The core principle of AVERI — that AI companies should not be permitted to evaluate their own safety without external verification (“AI companies shouldn’t be allowed to grade their own homework”) — extends the logic of financial auditing, pharmaceutical safety testing, and nuclear verification regimes to AI. Brundage has argued that one of the critical lessons from his OpenAI tenure is that “companies are figuring out the norms of this kind of thing on their own,” and that a universally applied external auditing standard is necessary to close the credibility gap between claimed and actual safety.

Key Contributions

  • “The Malicious Use of Artificial Intelligence” (2018) — Co-lead author of the first comprehensive framework for AI security threats; engaged 26 researchers across academia, civil society, and industry; shaped legislative and regulatory discussions in the UK, EU, and US.

  • OpenAI’s external red-teaming program — Initiated and institutionalized structured adversarial pre-deployment evaluation at OpenAI; helped establish the practice as a field-wide norm.

  • OpenAI System Cards — Led production of the first several system cards accompanying major model releases; defined the transparency reporting genre for frontier AI.

  • Compute Governance — Co-authored foundational research (arXiv:2402.08797) articulating why computing hardware is the most tractable point of AI governance leverage and what governance architectures might be built around it.

  • AVERI — Founded and leads the first nonprofit institution dedicated specifically to external AI auditing standards; the January 2026 launch paper co-authored with 30+ researchers provided a detailed operational framework for independent AI audits.

  • AI Governance Career and Ecosystem Building — Brundage’s widely-read FAQ on AI policy careers, his Substack, and his extensive public engagement have made him one of the field’s most cited practitioners for those entering AI governance work.

  • AGI Readiness framework — Led OpenAI’s internal work on the “levels of AI” framework and on structured assessment of both organizational and societal preparedness for increasingly capable AI — influencing how the frontier AI community discusses capability milestones.

Institutional Positions & Recognition

  • Institute for Progress — Non-Resident Senior Fellow; focuses on government acceleration of beneficial AI and societal resilience.
  • Center for a New American Security (CNAS) — Member, AI Governance Forum.
  • Institute for Advanced Study (IAS), Princeton — Member, AI Policy and Governance Working Group.
  • Epoch AI — Advisor; reflects his interest in rigorous quantitative assessment of AI progress trends.
  • RAND Corporation — Advisor; reflects engagement with national security dimensions of AI governance.
  • Axon Ethics Board (2018–2022) — Member until collective resignation over Taser drone proposal.
  • NSF Graduate Research Support — Research funded by National Science Foundation, Bipartisan Policy Center, and Future of Life Institute during PhD.

Key Relationships

  • David Guston — PhD advisor at Arizona State; leading scholar of responsible innovation and anticipatory governance; the theoretical framework Guston developed for thinking about governance of emerging technologies shapes Brundage’s approach to AI policy.
  • Joanna Bryson — PhD committee member and longtime collaborator; Bryson’s work on AI ethics and governance has been a continuing intellectual reference point.
  • Dario Amodei — Co-author of “The Malicious Use of AI” report; now CEO of Anthropic; Brundage’s 2018 work with Amodei at FHI/OpenAI represents an early intersection of the safety-focused researchers who would later diverge between OpenAI and Anthropic.
  • Jack Clark — Co-author of the 2018 malicious use report; later co-founded Anthropic with Amodei; another early collaborator from the Oxford/OpenAI safety community.
  • Sam Altman — Acknowledged in Brundage’s departure post; OpenAI CEO during Brundage’s tenure; thanked by Brundage for “trusting me with increasing responsibilities.”
  • Helen Toner — Co-author of the 2018 report; later a contentious OpenAI board member during the November 2023 crisis; shared early interest in AI governance from the FHI network.

Personal Style

Brundage is among the most intellectually transparent practitioners in AI policy: his departure post from OpenAI ran to several thousand words of self-analysis, distinguishing between his reasons for leaving in careful order and explicitly flagging the ways his own independence might be questioned. This reflexivity — about bias, about institutional interests, about the difference between research and advocacy — is consistent with his academic training in science and technology studies, which is oriented toward examining the conditions under which knowledge claims are made rather than just the claims themselves. His public engagement has a distinctive dual register: on technical policy questions (compute governance, auditing frameworks, frontier AI regulation) he publishes structured academic-style arguments; on broader questions of AI progress and urgency he writes more personally and is willing to make strong claims about the inadequacy of current governance. His Digg vibe profile captures this well: “Informing” and “Humorous” are roughly tied at the top, followed closely by “Teaching” and “Provocative” — a researcher who takes his subject seriously without taking himself too seriously.

References