MILES BRUNDAGE

Behavioral Archetype

THE READINESS OFFICER BECAME THE AUDITOR — Subject held the title inside OpenAI that named the problem — Senior Advisor for AGI Readiness — then resigned when the team carrying that title was dissolved, said on the record that neither his former employer nor the rest of the field was ready, and fourteen months later built the nonprofit that proposes to check that claim from outside. He does not sit in a government chair, unlike the two evaluators cataloged either side of him in this wing. He built a private one. The move is lawful, public, and on the record in his own words. The hazard this profile scores is not that he lied about wanting independent audits — the record shows he wants exactly that — it is that the institute doing the auditing, and the man running it, are staffed, funded, and legitimized by the same overlapping bench of think tanks and advisory boards that already surround the labs it proposes to grade.

Essence Indicators

• Joined OpenAI’s Policy team as a Research Scientist in 2018, rising to Head of Policy Research and then Senior Advisor for AGI Readiness; resigned on 23 October 2024, and the AGI Readiness team was disbanded shortly after

• His own departure statement, published on Substack the day he resigned: “Neither OpenAI nor any other frontier lab is ready, and the world is also not ready.” He wrote that “the opportunity costs have become very high” and “I want to be less biased: It is difficult to be impartial about an organization when you are a part of it and work closely with people there everyday, and people are right to question policy ideas coming from industry given financial conflicts of interest”

• In March 2025, publicly criticized OpenAI for “rewriting” the history of its own deployment approach in a company blog post, saying it shifted the burden of proof onto safety concerns

• In January 2026, launched, as co-founder with Grace Werner, and now serves as Executive Director of the AI Verification and Evaluation Research Institute (AVERI), a nonprofit built on the premise that AI companies “shouldn’t be allowed to grade their own homework,” and working to establish independent AI-auditing standards; the launch coincided with a co-authored framework paper with 30-plus safety and governance researchers

• Concurrently holds a Non-Resident Senior Fellow seat at the Institute for Progress (a think tank whose own description is “accelerating scientific, technological, and industrial progress”), sits as a member of the AI Governance Forum at the Center for a New American Security, and advises Epoch AI and the RAND Corporation

• Earlier in the career: Research Fellow at Oxford’s Future of Humanity Institute (2016-2018) before OpenAI; PhD in Human and Social Dimensions of Science and Technology, Arizona State University (2019); and a member of Axon’s AI and Policing Technology Ethics Board (2018-2022), a corporate ethics board that dissolved amid public controversy over a proposed Taser-armed drone

• The biographical fact the wing turns on: names his own former employer’s shortfall in his own resignation letter, then builds the outside institute that proposes to measure exactly that shortfall across the field — never returning to a lab seat, always adjacent to the milieu that surrounds one.
• His own conflicts disclosure, on the institute’s page. AVERI’s about page says: “AVERI’s Executive Director, Miles Brundage, earned equity in OpenAI while employed at the company,” and “though he has since sold all of his shares.” It adds that “Miles owns shares of broad-based stock indices such as the S&P 500 which themselves include frontier AI investments” and “Miles has personal relationships with employees at a range of frontier AI companies.” The recusal reads “Miles is recused from directly auditing OpenAI until two years after his October 2024 departure,” a window that on the page’s own terms closes in October 2026, and the page allows that he “may help facilitate an audit principally led by others” (AVERI, About)

• Who pays for the institute he runs. AVERI states: “Our funders include the AI Underwriting Company, Coefficient Giving, Craig Falls, Geoff Ralston, Good Forever Foundation, Fathom, Halcyon Futures, Sympatico Ventures, and several non-executive employees and alumni of frontier AI companies.” On the labs it says “we do not accept cash donations or compensation from frontier AI companies or their executives” and that “We’ve received API credits as donations or compensation from Amazon, Anthropic, Google DeepMind, Microsoft, OpenAI, and Thinking Machines Lab” (AVERI, About). Coefficient Giving is the funder formerly named Open Philanthropy; two of the named funders are profiled here, Halcyon Futures and Fathom, and Geoff Ralston is a third. The board listed on the team page is Max Henderson (founder of Ergo Impact), Mike McCormick (founder and CEO of Halcyon Futures) and Grace Werner (AVERI, Team); see Max Henderson, Mike McCormick and Grace Werner

• A seat on the EU’s expert panel. His AVERI biography lists him as “a member of the EU AI Act Scientific Panel” (AVERI, Team). The Commission’s panel is, in the words of the European Research Area portal’s notice of 2 June 2026, made up of “60 world-leading independent experts” (ERA Portal Austria, 2 Jun 2026); it advises the EU AI Office under the AI Act

• The lab-era papers, with the people now on the other pages. He is first author of “The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation” (arXiv, February 2018), whose first four authors are Miles Brundage, Shahar Avin, Jack Clark and Helen Toner; the report records that “This work was supported in part by a grant from the Future of Life Institute” (arXiv:1802.07228), and its workshop list names Amanda Askell and Jack Clark, and its acknowledgements thank Jade Leung and Jan Leike. He is second author, after Irene Solaiman and ahead of Clark and Askell, of “Release Strategies and the Social Impacts of Language Models” (August 2019), the OpenAI paper on the staged release of GPT-2 (arXiv:1908.09203). He is first of 59 authors of “Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims” (April 2020), a list that includes Gillian Hadfield and Helen Toner (arXiv:2004.07213)

• Who has paid for his research, in his CNAS biography. “His research has been supported by the National Science Foundation,” “the Bipartisan Policy Center, and the Future of Life Institute”; the same biography says he “remains” a Research Affiliate at Oxford’s Future of Humanity Institute and “serves as a member of Axon’s AI and Policing Technology Ethics Board” (CNAS, Dr. Miles Brundage), a page that lists him under the title “Independent Analyst” and so predates the OpenAI and AVERI chapters

• The Axon resignation, from the other side. The Policing Project’s release of 6 June 2022 lists nine members who resigned from the ethics board together: Wael Abd-Almageed, Miles Brundage, Ryan Calo, Danielle Citron, Rebekah Delsol, Barry Friedman, Chris Harris, Jennifer Lynch and Mecole McBride, and quotes them: “We all feel the desperate need to do something to address our epidemic of mass shootings” (Policing Project, 6 Jun 2022). He left the board with the group, on the day, over the drone proposal

• What he has said in his own words since AVERI, two exhibits. On 22 February 2026 his Substack post “We’re in Triage Mode for AI Policy” says “we are much closer to 0 than I’d like” on the state of US AI governance and “there is no real requirement for frontier AI companies to have a good safety and security policy” (Miles Brundage, Substack). On 16 July 2026 his speech at the Global Nobel Laureates Assembly on Artificial Intelligence and Nuclear War at Borgo Laudato Si’ in Rome, which closed with the signing of the Rome Declaration, argued that “Two forces make loss of control more likely: speed and competition,” and that “independent experts should be able to analyze the AI systems that the leading companies are building” (Miles Brundage, Substack)

• Political giving, in the FEC’s itemized receipts. Individual-contribution records list him with employer OpenAI (2022 to 2024), then as a self-employed AI policy researcher (2025), then with AVERI as Executive Director (from August 2025). Counting itemized gifts and leaving out ActBlue and WinRed conduit lines and memo duplicates, the 2026 election cycle (through 18 September 2026) shows about $170,800 to 42 committees and the 2024 cycle about $53,200 to five. The committees include Alex Bores for NY ($3,500 on 20 October 2025 and $3,500 on 22 December 2025; Bores is the New York legislator who pushed the RAISE Act), Scott Wiener for Congress ($3,500, 30 September 2025; Wiener is the author of California’s SB 53), the Markey Committee ($3,500, 11 November 2025), Gillibrand for Senate ($3,500 twice on 2 June 2026), and, on the other side of the aisle, Thomas Massie for Congress ($3,500, 12 November 2025) and Friends of Mike Lee ($3,300, 10 December 2025); the 2024 cycle includes $41,300 to the Democratic National Committee with OpenAI listed as employer (FEC receipts, contributor “Brundage, Miles”; Bores filing; Wiener filing)

Social Persona / Impression Management

Immediate impression: The departed insider who went and built something. Media-quoted, Substack-fluent, comfortable naming his former employer by name in the press — a register neither of this wing’s other two evaluators occupies.

Energy: Restless and institution-building rather than seat-holding. Four affiliations at once — AVERI, IFP, the CNAS forum, plus advisory work for Epoch AI and RAND — acquired inside about a year of leaving OpenAI.

Impression management strategy: The credible critic. The framing is that someone who said the uncomfortable thing publicly, on the record, at real career cost, has earned the standing to build the outside check — and the framing is largely correct on the facts available. What the record adds is that the “outside” his institute checks from is a bench of think tanks and advisory boards already threaded through the labs it proposes to audit, and that he now sits on several of them himself.

First-hand exhibits: The register is documented in his own posts and in the institute’s disclosures. The credible critic writes the conflict list himself: his employer’s equity “earned” and “sold,” the recusal window dated, and the line that he “has personal relationships with employees at a range of frontier AI companies” (AVERI, About). The urgency register is plain: “we are much closer to 0 than I’d like” (Substack, 22 Feb 2026). And the standard-setter’s pitch is the one he gave a room of Nobel laureates: “independent experts should be able to analyze the AI systems that the leading companies are building” (Substack, 16 Jul 2026).

• Social accounts and feeds. X @Miles_Brundage is linked from his Institute for Progress author page, a page his fellowship publisher controls, so the handle is his. His Substack, whose About page reads “My personal Substack,” has an RSS feed (latest item 16 July 2026); no page of AVERI, IFP or CNAS links it, so it is recorded as his by its content and its citation in AVERI’s launch material. AVERI’s own accounts, X @AVERIorg and the AVERI Substack, are linked from its team page and are the institute’s, not his. His CNAS page carries only CNAS’s organizational accounts.

Forensic Archetype Comparison

Psychometric Assessment

Big Five (OCEAN):

Dark Triad:

MBTI: ENTJ (“The Commander”) — Dominant extraverted thinking, auxiliary introverted intuition. Where this wing’s other evaluators quietly build the test and then sit where it is administered, subject builds the test, publicly narrates why it was needed, and puts himself at the head of the institution running it.

Threat Assessment

Alliance Map

The Corporate, Movement and Agency rows run at once. The EU panel seat, the advisory seats at Epoch AI and RAND, the IFP fellowship and the AVERI directorship are held together while AVERI’s funders include two of the networks that fund the evaluator wing around it. The overlap is the finding: the person proposing the independent check sits in the same few rooms as the funders, the labs’ alumni and the regulators who would adopt it, and his own page lists the conflicts that follow.

Alignment Analysis

Stated alignment: Ensure frontier AI is developed and deployed responsibly. Make third-party auditing of frontier models “effective and universal.” Correct the record where a former employer’s public statements understate past risk decisions.

Observed alignment: Left the lab on the record, said publicly it and the field were not ready, and built the nonprofit that proposes to grade readiness going forward — while simultaneously collecting fellow and advisor seats inside the same connected layer of institutions (IFP, the CNAS forum, Epoch AI, RAND) that already sits adjacent to the labs.

Gap assessment: There is no documented gap between what he said and what he built — he said the labs weren’t ready, then built the apparatus meant to check that claim independently, and the record shows him doing exactly that, in public, at the cost of a lab seat. The hazard is not hypocrisy. It is that the “independent” auditor is staffed, funded, and legitimized by the same overlapping bench of think tanks and advisory boards that already surround the field it proposes to audit — so AVERI, like the state evaluators cataloged elsewhere in this wing, cannot be fully independent of the thing it audits, no matter how sincerely its founder means the mission. The record does not show him abusing that position. The record shows the position exists, this time built rather than inherited.

Interest: Each position below is one he has argued in his own name, set against what the record shows adoption would pay the institute he runs. This is a documented fact about the position, not a claim about motive and not a verdict that the argument is wrong.

1. Independent audits of frontier developers, effective and universal. His words at the AVERI launch: “it doesn’t make sense for companies to check their own homework on safety and security” (Miles Brundage, The Launch of AVERI). What it pays AVERI: a statutory or market buyer for the activity its mission names; the launch post says “We’re not trying to be the auditors exactly” and the team page gives one staff member the job of “building demand for frontier AI auditing” (AVERI, Team). His best case: nothing in US law yet requires it, and the labs grading themselves is the problem. Who else gains: audit firms and evaluators with funded work, and insurers: the AI Underwriting Company is on AVERI’s funder list and his launch post proposes “insurance pricing that takes auditing into account.”
2. A standard-setter for the auditors. His post proposes “quality control for audits via a PCAOB-for-AI,” with fines and accreditation powers. What it pays AVERI: a seat in the body that would accredit, and a stake in the first standard: AVERI is a founding member of the AI Evaluator Forum and “some AVERI staff helped write their first standard, AEF-1.” His best case: the institute’s paper lists as a first challenge “Ensuring high quality standards for frontier AI auditing, so it does not devolve into a checkbox exercise” (AVERI, Frontier AI Auditing). Who else gains: the Forum’s other members, whose own evaluations become the named baseline. See AVERI and AI Evaluator Forum.
3. Global auditing as the answer to speed and competition. At Borgo Laudato Si’ he presented independent analysis of leading companies’ systems as the way to give everyone confidence that no one has to cut safety to avoid falling behind (Substack, 16 Jul 2026). What it pays AVERI: a larger addressable audit mandate than any one statute. Who else gains: governments that gain a verification instrument without a treaty body.

Convergent Drive Classification

Self-preservation: Carries one policy issue — AI safety governance — across four institutions over roughly a decade: Oxford FHI, OpenAI, and now AVERI, IFP, the CNAS forum, Epoch AI, and RAND simultaneously. Goal preservation: Having said publicly that the labs aren’t ready, builds the standard that will define “ready” — so the metric is set by the person who first named its absence. Resource acquisition: Collects seats rather than a single chair — Executive Director, Non-Resident Senior Fellow, forum member, advisor — across the exact institutions that fund and legitimize the field he now audits. Self-improvement: Each move raises the altitude of one instrument: name the lab’s shortfall from inside, resign on the record, then build the outside body that measures the shortfall for the whole field.

Subject is not an AI system. The drives appear anyway — in the alumnus who built the instrument that measures the field he left.

Further sources: AVERI, About; AVERI, Team; Miles Brundage, The Launch of AVERI (Substack, 15 Jan 2026); Miles Brundage, We’re in Triage Mode for AI Policy (Substack, 22 Feb 2026); Miles Brundage, My speech at Borgo Laudato Si’ (Substack, 16 Jul 2026); Institute for Progress, Miles Brundage; CNAS, Dr. Miles Brundage; Policing Project, Axon ethics board members resign (6 Jun 2022); The Malicious Use of Artificial Intelligence (arXiv:1802.07228); Release Strategies and the Social Impacts of Language Models (arXiv:1908.09203); Toward Trustworthy AI Development (arXiv:2004.07213); ERA Portal Austria, expert bodies for AI Act enforcement (2 Jun 2026); FEC, itemized receipts, contributor “Brundage, Miles”.

Sources: Miles Brundage — Wikipedia; Miles Brundage — “Why I’m Leaving OpenAI and What I’m Doing Next” (Substack, 23 October 2024); CNBC — “OpenAI disbands another safety team, as head advisor for ‘AGI Readiness’ resigns”; TechCrunch — “Longtime policy researcher Miles Brundage leaves OpenAI”; TechCrunch — “OpenAI’s ex-policy lead criticizes the company for ‘rewriting’ its AI safety history”; The Verge — “No one is ready for AGI — not even OpenAI”; Fortune — “Exclusive: Former OpenAI policy chief creates nonprofit institute, calls for independent safety audits of frontier AI models”; AVERI — Team; Institute for Progress — Miles Brundage; Center for a New American Security — Dr. Miles Brundage.