LUKE MUEHLHAUSER

Behavioral Archetype

THE ALLOCATOR — Subject is the man who decides, inside the field’s largest funder, which AI-governance bodies, think tanks and evaluators get money. He does not train a model, set a model’s refusals or run a lab. He recommends grants, and the recipients (RAND, Georgetown’s CSET, the Wilson Center, Epoch AI) are the institutions that tell governments what advanced AI is and what to do about it. Before that he ran the research institute the movement began with, and until May 2024 he sat on the board of a lab that, per POLITICO, his funder’s principal donor had backed. The throughline is not a title. It is that one career runs from the movement’s first institute, through the funder, to a seat on a lab’s board and back out of it. This profile scores the seat and the position, not the man.

Essence Indicators

• Current role, from his funder’s own pages. Coefficient Giving’s team page gives his title as “Managing Director, AI Governance & International Policy” and says “Luke joined Coefficient Giving in 2015. He leads our grantmaking on AI governance and international policy.” (Coefficient Giving, Luke Muehlhauser) The Navigating Transformative AI fund page lists him as a Managing Director beside Claire Zabel, names Good Ventures as its partner, and states “570+ grants made” (Coefficient Giving, Navigating Transformative AI). His title appears as “Managing Director, AI Governance & Policy” in Epoch’s January 2026 impact report, and as “program director for AI governance and policy” in the funder’s October 2025 strategy essay (Epoch AI; Coefficient Giving, Our Approach to AI Safety and Security).

• How he got there. The team page says that before joining he “was Executive Director of the Machine Intelligence Research Institute in Berkeley, California,” and that as a Research Analyst and then Senior Research Analyst his work “spanned a wide range of topics, particularly the social sciences, global catastrophic risks, animal consciousness, and forecasting.” (Coefficient Giving) His own farewell post to MIRI, dated May 6 2015: “I’m excited to have accepted a research position at GiveWell.” and “I look forward to supporting MIRI going forward, both financially and as a close advisor.” (Luke Muehlhauser, MIRI, May 6 2015) The funder’s own page dates its AI safety work to 2015 (“began supporting AI safety and security work in 2015”), the year he arrived (Coefficient Giving).

• Author of the funder’s AI-governance priorities. On December 8 2020 he published “Our AI Governance Grantmaking So Far”: “In this post, I aim to clarify our priorities in AI governance, and summarize our AI governance grantmaking so far.” The grantees he names for the priorities are GovAI, CSET and the Wilson Center. (Luke Muehlhauser, Open Philanthropy (now Coefficient Giving), Dec 8 2020)
• Grants he recommended, from the funder’s own grant pages. Grant pages name him as grant investigator for the Wilson Center’s AI policy training program ($291,214, May 2021), a Rethink Priorities AI governance research contract ($612,185, December 2021) and Georgetown CSET general support (August 2021). RAND gifts “recommended by Luke Muehlhauser, who leads our grantmaking on AI governance” are listed at $10.5 million (page dated January 2024), $2.5 million (August 2024), $5 million (November 2024) and $2.5 million (July 2025), each “to be spent at the discretion of RAND president and CEO Jason Matheny” except the August 2024 page, which says “to be spent at RAND President Jason Matheny’s discretion.” (Coefficient Giving, grant pages listing Muehlhauser as investigator; Coefficient Giving, RAND gifts) The funder’s September 9 2026 announcement says “CG grantmakers Catherine Brewer and Luke Muehlhauser recommended a $68 million renewal over 3 years, more than double our previous four years of support” for Epoch AI. (Emily Oehlsen, Coefficient Giving, Sep 9 2026)

• A seat on a lab’s board, and the exit. POLITICO reported on October 13 2023 that he was “one of Anthropic’s four board members” and that the funder’s spokesperson said he “holds no financial interest” in the company. (POLITICO, Oct 13 2023, as reported) Anthropic’s own announcement of May 29 2024 says “Luke Muehlhauser has decided to step down from his Board role to focus on his work at Open Philanthropy.” and that “His guidance and thoughtfulness helped the company navigate our first phase of growth.” (Anthropic, May 29 2024) Skype co-founder Jaan Tallinn, an investor in Anthropic’s 2021 round, told Semafor that “Instead of taking a board seat, Tallinn said he wanted someone he trusted to join Anthropic’s board” and “argued for Luke Muehlhauser” (Semafor, Apr 28 2023, as reported by Semafor).

• He is a consumer of the evaluators he funds. In Epoch’s own impact survey he wrote: “A key crux for CG’s AI grantmaking strategy is AI timelines.” and that “Epoch reports have been a frequent source of parameter value estimates.” (Epoch AI, 2025 impact report, Jan 16 2026) The funder that pays for the measurement uses the measurement to set its strategy. That is a documented act of the program, stated here as the structure, not as a fault.
• Author and public writing. His book site describes him as “the former Executive Director of the Machine Intelligence Research Institute” and author of the intelligence-explosion book (Facing the Intelligence Explosion, About). His personal domain did not answer, and no personal blog, X, Mastodon or other account is linked from his Coefficient Giving page.

Social Persona / Impression Management

Immediate impression: The careful analyst. His public register is the long investigative post: a definition, a list of grants, a stated uncertainty, a footnote giving the source for each claim.

Energy: Low-key, written, methodical. The record is posts, grant pages and a team bio, not keynotes. The funder’s CEO-level announcement describes him and two colleagues as having “been focused on these problems for many years, well before the rest of the world caught up.” (Coefficient Giving)

Impression management strategy: The honest broker of uncertainty. His 2020 post states the limits of the whole enterprise in the first person: “Would tighter regulation of AI technologies in the U.S. and Europe meaningfully reduce catastrophic risks, or would it increase them” (Coefficient Giving, Dec 8 2020), and describes the field he funds as one where “relatively few people who work on AI governance share our focus on improving likely outcomes from transformative AI.” He leaves the Anthropic board when the funder’s policy work and a lab seat sit awkwardly together, and says so in one line through the lab’s announcement. Whether the candor is temperament or positioning is not establishable from the outside, and for the allocator it never needs to be.

Forensic Archetype Comparison

Psychometric Assessment

Big Five (OCEAN): observations of the documented public role, not claims about private character.

Dark Triad:

MBTI: INTP (“The Logician”) — Dominant introverted thinking, auxiliary extraverted intuition. A definer-of-terms and enumerator of uncertainties, whose public work is analysis for a decision rather than advocacy of one.

Threat Assessment

Score basis. Atk 7, def 8, hp 7 at hazard 72. Atk is allocation leverage: the funder’s team page says he leads its AI governance and international policy grantmaking, and its grant pages credit him as the recommender on named RAND, CSET, Wilson Center, Rethink Priorities and Epoch grants (Coefficient Giving; Coefficient Giving). It is a recommendation seat inside a funder with a president and a principal donor, which is why it sits below the deployment authority of a lab leader. Def is insulation: the fund’s partner is Good Ventures and none of the grantees pays the funder (Coefficient Giving); he left the Anthropic board in May 2024 (Anthropic). Hp is durability: “570+ grants made,” multi-year renewals and the 2020 priorities that the funder’s own 2025 essay still cites outlast any one seat (Coefficient Giving). Peers: Ajeya Cotra (5/8/7, hazard 64) led the same funder’s technical AI safety grantmaking in 2024 and is now at METR, so her current allocation leverage is lower while def and hp match; Jade Leung (7/9/8, hazard 76) shares atk but holds a statutory-adjacent government seat with greater insulation; Helen Toner (6/8/8, hazard 80), a former analyst at the same funder, leads an institute the funder paid, with hp higher for the institute she runs. Holden Karnofsky (8/9/8, hazard 82), the funder’s co-founder, is a point higher on every component for the founder’s insulation and the institutions he built.

Alignment Analysis

Stated alignment: Reduce the risk of catastrophe from advanced AI and help society prepare for it. In the fund’s own words, “AI governance and policy work to develop frameworks for safe, secure, and responsibly managed AI development.” (Coefficient Giving)

Observed alignment: Decide, inside a single large funder, which think tanks, training programs and evaluators carry AI-governance work: RAND and its Pardee RAND Graduate School program, Georgetown CSET, the Wilson Center, Rethink Priorities, Epoch. Sit on the board of a lab that, per POLITICO, his funder’s principal donor had backed, then step out of it.

Gap assessment: The stated and observed alignments overlap wherever “fund governance work” and “fund the institutions whose view of governance the funder already shares” are the same act, and his own 2020 post says the second is close to the whole program: the grantmaking “tends to focus on” research, advocacy and training aimed at “intermediate goals that we’ve come to think will improve expected transformative AI outcomes.” The uncertainty is stated in his own words, including whether tighter regulation would “reduce catastrophic risks, or would it increase them”. The record does not settle whether the portfolio is service or positioning, and for the allocator it never needs to.

Interest: What each position would pay the institutions he funds or the seats he held, documented, with his best case.

• Fund governance training and advice for governments. Position, from his 2020 post: the grantmaking aims at “better-informed AI governance training and advice for governments, companies, and other actors,” with grants to CSET and the Wilson Center as examples (Coefficient Giving). What it pays: the funder’s pages list RAND’s technology policy training program and the Pardee RAND Graduate School among the uses of four gifts he recommended (Coefficient Giving); a government that wants trained staff and advice buys from those same programs. His best case, from the same post: “there has been substantial AI governance work that I suspect has increased the odds of good outcomes from transformative AI,” and “relatively few people who work on AI governance share our focus on improving likely outcomes from transformative AI,” so the funder pays for the focus the field would otherwise lack. Who else gains: the think tanks and graduate programs that train and place staff, and any agency that hires the graduates.
• Better measurement of dangerous capabilities. Position, on the program’s capability-evaluation request for proposals: “We are worried that the current evaluations paradigm isn’t mature enough to play this role.” and “Governments, regulators, and society at large need robust, reliable measurements to understand and respond appropriately to advances in AI capabilities.” (Coefficient Giving, RFP: Improving Capability Evaluations) The page says applicants who miss it may submit to the AI Governance and Policy General RFP, which is his program. What it pays: Epoch AI’s three-year $68 million renewal that he recommended with a colleague, and the evaluators and benchmark builders the page says it wants to fund; the page notes that “multiple benchmarks funded through that RFP have already been used in pre-deployment testing of frontier models.” His best case is the page’s: capability evaluation underlies the if-then commitments governments and labs now write. Who else gains: the labs, whose pre-deployment testing gains outside benchmarks, and the AI safety institutes that run them.
• Calibrated caution on regulation. Position, in his words: it is hard to know which intermediate goals help, and “tighter regulation of AI technologies in the U.S. and Europe” might “reduce catastrophic risks” or increase them. What it pays: the documented portfolio is research and training institutions (RAND, CSET, Epoch, Rethink Priorities) rather than rule-advocacy groups; the record documents no rule he advocates that would bind one party and exempt another. His best case: the sign-uncertainty argument itself. Who else gains: any actor facing a proposed rule, which gains time while the research continues.

Interest is a documented fact about each position, not a claim about motive, and not a verdict that the argument is wrong. The best case is stated at the strength he states it.

Alliance Map

Several of these run at the same time. Through May 2024 he recommended grants to evaluators and policy bodies while sitting on the board of a lab that those evaluators grade and those bodies advise governments about; per POLITICO the funder’s principal donor was also a lab backer. In 2026 he leads the AI governance grantmaking at the field’s largest funder, recommends the renewal of an evaluator whose reports his own program uses to set strategy, and is one of a small circle of former and current funder staff (Cotra, Toner, Karnofsky) now spread across METR, CSET, Anthropic and Epoch. The overlap is the finding, not a coordination claim.

Convergent Drive Classification

Self-preservation: Survives every institutional transition by carrying the working method, not the title. Executive director, research analyst, program lead, board member: one network. Goal preservation: Defines the portfolio’s frame in his own priorities post, so the goal (“improving likely outcomes from transformative AI”) is protected by the grant list before it is ever debated. Resource acquisition: Trades in the scarce resource of the field: funding for the institutions that brief governments and test models, allocated by recommendation rather than ownership. Self-improvement: Each role is a step up in altitude on the same question: the institute, the funder’s research staff, the funder’s AI governance program, a lab’s board.

Subject is not an AI system. The drives appear anyway — in the allocator whose product is the portfolio of institutions that govern and measure the systems.

Funding record: the grants, board seats and fellowships behind this profile are documented on The EA to AI-Safety Funding Pipeline and the evaluator alliances map.

Sources: Luke Muehlhauser — Wikipedia; Coefficient Giving — Luke Muehlhauser; Coefficient Giving — Navigating Transformative AI; Coefficient Giving — Our AI Governance Grantmaking So Far (Dec 8 2020); Coefficient Giving — Our Approach to AI Safety and Security; Coefficient Giving — We’re Urgently Scaling Our Work on AI and Biosecurity (Sep 9 2026); Coefficient Giving — RFP: Improving Capability Evaluations; Coefficient Giving — grant pages (RAND); Coefficient Giving — grant pages (investigator credits); Anthropic — Jay Kreps appointed to board (May 29 2024); MIRI — A fond farewell and a new Executive Director (May 6 2015); Epoch AI — 2025 impact report; Facing the Intelligence Explosion — About; POLITICO — How a billionaire-backed network of AI advisers took over Washington, Oct 13 2023; Semafor — Apr 28 2023.