Miles Brundage
Last reviewed
May 31, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 ยท 2,074 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 31, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 ยท 2,074 words
Add missing citations, update stale details, or suggest a clearer explanation.
Miles Brundage is an American researcher who works on the policy and governance of artificial intelligence. He spent more than six years at OpenAI, where he led policy research and later served as senior advisor for AGI Readiness, a role focused on whether the company and the wider world were prepared for artificial general intelligence. He left OpenAI in October 2024, stating in a widely covered note that neither OpenAI nor any other frontier lab was ready for advanced AI and that he could be more useful working independently. He is a co-author of influential reports including "The Malicious Use of Artificial Intelligence" (2018) and "Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims" (2020), and he helped shape the staged release of OpenAI's GPT-2 language model. In January 2026 he launched a nonprofit, the AI Verification and Evaluation Research Institute (AVERI), to promote independent auditing of frontier AI systems.
| Field | Detail |
|---|---|
| Full name | Miles Brundage |
| Occupation | AI policy researcher |
| Known for | AI governance, AI safety policy, AGI readiness, AI auditing |
| Education | B.A. Political Science, George Washington University (2010); Ph.D. Human and Social Dimensions of Science and Technology, Arizona State University (2019) |
| Notable employer | OpenAI (2018 to 2024) |
| Roles at OpenAI | Research scientist on the policy team, head of policy research, senior advisor for AGI Readiness |
| Earlier affiliation | Research Fellow, Future of Humanity Institute, University of Oxford (2016 to 2018) |
| Departed OpenAI | October 2024 |
| Current work | Founder and lead of the AI Verification and Evaluation Research Institute (AVERI), launched January 2026 |
| Other roles | Non-Resident Senior Fellow, Institute for Progress; advisor to Epoch AI and RAND; member of Axon's AI and Policing Technology Ethics Board (2018 to 2022) |
| Key publications | "The Malicious Use of Artificial Intelligence" (2018); "Release Strategies and the Social Impacts of Language Models" (2019); "Toward Trustworthy AI Development" (2020) |
Brundage earned a bachelor's degree in political science from George Washington University in 2010. [1] After graduating he worked for about two years at the Advanced Research Projects Agency-Energy (ARPA-E), part of the United States Department of Energy, and he held an internship at the Institute for Human and Machine Cognition. [1][2]
He then moved into academic study of emerging technology. He completed a Ph.D. in Human and Social Dimensions of Science and Technology at Arizona State University in 2019, a program housed in the university's School for the Future of Innovation in Society. [1][3] His doctoral work centered on how societies anticipate and govern fast-moving technologies, with AI as a primary case. [3]
From 2016 to 2018 Brundage was a Research Fellow at the Future of Humanity Institute at the University of Oxford, where he was part of the group that later became the Centre for the Governance of AI. [1][4] His research there looked at methods for analyzing AI development scenarios and the policy responses they might call for. [4] He has continued to hold a research affiliate or analyst status with the Center for a New American Security (CNAS). [2]
Brundage's published work sits at the intersection of AI safety and AI governance. Several of his reports have been cited widely in academic and policy discussions about how advanced AI should be developed and released.
In February 2018 Brundage was the lead author of "The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation," a report written with Shahar Avin, Jack Clark, and more than twenty other researchers drawn from universities, civil society groups, and industry. [5][6] The report surveyed how AI capabilities could expand security threats across digital, physical, and political domains, including automated cyberattacks, the use of drones, and synthetic media used for disinformation. It set out four high-level recommendations, among them closer collaboration between policymakers and researchers and more attention to the dual-use nature of AI research. [5][6] The document is often described as one of the earlier coordinated attempts to map the security risks of modern AI.
Brundage became closely associated with debates over how AI labs should release powerful models. When OpenAI built GPT-2 in 2019, the organization chose not to publish the full 1.5 billion parameter model immediately, citing concerns about misuse such as mass-produced misleading text. [7][8] Instead it used a staged release, putting out a small 124 million parameter version in February 2019, a 355 million parameter version in May, a 774 million parameter version in August, and the full model in November. [8]
Brundage was a co-author of the August 2019 report "Release Strategies and the Social Impacts of Language Models," written with Irene Solaiman, Jack Clark, Amanda Askell, and others, which explained the reasoning behind the staged approach and the tradeoffs of different release strategies. [7][8] The episode prompted broad discussion about whether withholding model weights was a useful safeguard or an obstacle to open research, and the term "release norms" entered common use in the field partly through this work.
In April 2020 Brundage was a lead author, again with Shahar Avin and a large group of co-authors, of "Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims." [9] The report argued that developers need ways to make claims about the safety, security, fairness, and privacy of their systems that outside parties can actually check. It analyzed ten mechanisms spanning institutions, software, and hardware, including third-party auditing, red-teaming exercises, bias and safety bounties, and audit trails. [9] This focus on external verification later became the organizing idea of his independent work.
Brundage joined OpenAI in 2018. [1][10] He started as a research scientist on the policy team and went on to become the organization's head of policy research, working in particular on the responsible deployment of language generation systems. [1][2] In his later years at the company he held the title of senior advisor for AGI Readiness. [10][11]
The AGI Readiness team advised OpenAI's leadership on the company's capacity to handle increasingly capable AI and on whether society was prepared to manage systems that might match or exceed human ability at many tasks. [10][11] The mandate covered topics such as forecasting AI progress, frontier AI regulation, the economic effects of AI, and the safe deployment of new systems. [12]
Brundage announced his departure from OpenAI on October 23, 2024, in a post titled "Why I'm Leaving OpenAI and What I'm Doing Next." [10][12] By his own account he had been at the company for over six years. [12] OpenAI said at the time that it would disband the AGI Readiness team and distribute its members across other parts of the organization. The economic research work moved under the company's new chief economist, Ronnie Chatterji, and some remaining projects went to Joshua Achiam, OpenAI's head of mission alignment. [10][11]
In the note Brundage gave several stated reasons for leaving. He wrote that he wanted to publish more freely and work on research topics that he did not have time for inside the company, that it was difficult to be impartial about an organization while being part of it, and that he had completed much of what he set out to do in the AGI Readiness role. [12] He said he believed he could have more impact as a researcher and advocate in the nonprofit sector. [10][12]
The most quoted line from the post was his assessment of readiness for advanced AI. He wrote, "In short, neither OpenAI nor any other frontier lab is ready, and the world is also not ready." [12] He added that this view was not especially controversial among OpenAI's leadership and that the question was whether the company and the world would get ready in time. [12] Reporting on the departure placed it within a wider series of exits from OpenAI in 2024, which included chief technology officer Mira Murati and other senior staff. [11][13]
OpenAI responded by saying it supported his decision. A company spokesperson said, "We fully support Miles' decision to pursue his policy research outside industry and are deeply grateful for his contributions," and described his move to independent research as a chance to have an impact on a wider scale. [10][11]
After leaving OpenAI, Brundage continued to write and speak about AI governance, including on his Substack, and he took on advisory roles. He became a Non-Resident Senior Fellow at the Institute for Progress and an advisor to the research group Epoch AI and to the RAND Corporation. [1]
In January 2026 he announced the launch of the AI Verification and Evaluation Research Institute, known as AVERI, a nonprofit he had spent the previous year co-founding. [14][15] AVERI is set up as a United States 501(c)(3) organization, and its stated goal is to make third-party auditing of frontier AI effective and widely used. The institute defines this as rigorous external verification of AI developers' safety and security claims, and evaluation of their systems against relevant standards, based on secure access to information that is not public. [14][15] Brundage has summarized the underlying argument by saying that AI labs should not be left to "grade their own homework." [14]
AVERI says it aims to study what auditing standards should look like and to support policies that would push frontier developers toward external auditing, rather than to carry out audits itself. [14][15] Reporting at launch said the organization had raised about 7.5 million dollars toward a target of 13 million dollars, intended to support a staff of around fourteen, with funders including the former Y Combinator president Geoff Ralston and the AI Underwriting Company. [14][15] The launch coincided with a research paper, co-authored by Brundage and more than thirty researchers, that set out a framework of "AI Assurance Levels" describing increasingly demanding tiers of verification for AI systems. [14][15]
Brundage is frequently described in coverage of the field as one of the more prominent figures working on AI policy. [10][11] His reports on malicious use, release strategies, and verifiable claims are recurrent reference points in research and policy writing about how powerful AI should be built and shared. [5][7][9] His public departure from OpenAI, and his statement that no frontier lab was ready for advanced AI, drew wide attention because it came from someone who had worked inside one of the leading labs. [10][11][13] His later turn toward independent auditing, expressed through AVERI, reflects a broader argument in AI governance that external checks are needed alongside the internal safety work done by the companies themselves. [14][15]