EMILY OEHLSEN

Behavioral Archetype

THE ALLOCATOR WHO SHORTENED THE APPROVAL CHAIN — Subject is the second-in-command of the funder that Open Philanthropy became, and since March 2025 the person who runs its AI safety and biosecurity grantmaking. An economist by training who moved from running global health giving to running the AI line, she is the executive under whom the AI figure went from $168 million in 2024 to $351 million in 2025 to “on track to commit over $1 billion” this year, by her own account, and under whom, in her words, “a typical grant now needs active sign-off from only one person in addition to the primary grant investigator.” The hazard is the position: one seat that decides how fast the largest philanthropic checkbook in AI safety moves, which institutions grow on it, and how few people must agree. This profile scores the seat, not the woman.

Essence Indicators

• President, Coefficient Giving. The funder’s leadership page lists Alexander Berger as CEO, Cari Tuna as Chair and “Emily Oehlsen” as President; her team-page bio reads “Emily leads Coefficient Giving’s work on AI and biosecurity. She previously led our work on global health and wellbeing and farm animal welfare. Since joining cG, she has managed teams allocating over $2 billion in funding. She also helped launch Coefficient’s first collaborative fund, the Lead Exposure Action Fund” (Coefficient Giving, Team; Coefficient Giving, Emily Oehlsen).

• The route to the seat. Berger, February 1, 2024: she “joined Open Phil as a research fellow on the GHW Cause Prioritization team in April 2021, was promoted to co-run that team in March 2022, and then effectively started running all of GHW on an interim basis about a year ago”; she “became Managing Director for our work on Global Health and Wellbeing (GHW) in July 2023.” The funder’s March 6, 2025 progress post: “Emily Oehlsen was promoted to President,” and, of her new remit, that she “took on oversight of our Biosecurity & Pandemic Preparedness program in mid-2024 and is adding oversight of our AI work this month.” Berger’s 2025 letter: “Our President Emily Oehlsen, who oversees our work on global catastrophic risks, and Otis Reid … both joined Coefficient in 2021 as individual contributors and quickly took on much larger roles” (Berger, Feb 1, 2024; Coefficient Giving, Our Progress in 2024; Berger, 2025 Letter from the CEO).

• Before the funder. The bio: “Before joining cG in 2021, she worked at DeepMind, Uber, and TGG Group. She holds a B.S.F.S. in International Political Economy from Georgetown and a D.Phil. in Economics from Oxford.” Berger adds: “Prior to Open Phil, Emily worked on policy and research at DeepMind and Uber, and was a grad student in economics at Oxford” (Coefficient Giving, Emily Oehlsen; Berger, Feb 1, 2024).

• The September 9, 2026 post. “I began leading Coefficient Giving’s grantmaking on AI safety and biosecurity 18 months ago.” The numbers: “In 2024, Coefficient Giving committed $168 million across our technical AI safety, security, and fieldbuilding work.” Then: “In 2025, we more than doubled that to $351 million, and this year we’re on track to commit over $1 billion. In August alone, we committed $352 million across this portfolio.” Biosecurity: “from $30 million in 2024 and $65 million in 2025, to more than $150 million so far this year.” Her footnote defines the term: “‘Committed’ refers to grants we’ve formally recommended for funding” (Emily Oehlsen, Sep 9, 2026). Her reason: “transformative AI may arrive soon, and the world isn’t ready for it,” and “My own timelines have also shortened,” “largely in response to rapid progress in capabilities, the rise of agentic coding, unprecedented revenue growth at frontier AI companies, and my own day-to-day use of AI. I’m uncertain about these projections, and I think most of our work is worth doing even under long timelines.”

• What she changed, in her words. “The majority of the GCR grants over $20 million we’ve ever made have been made this calendar year”; “we recently approved $160 million for Resolution, our largest ever AI safety grant, to support a new organization led by Geoffrey Irving scaling rigorous alignment research”; “We’ve shortened grant investigations”; “a typical grant now needs active sign-off from only one person in addition to the primary grant investigator”; “we recently approved an initial grant of $10 million to help Fields Medal recipient Jacob Tsimerman establish an institute encouraging top mathematicians to work on AI safety; the grant was approved 12 days after we had the first conversation”; “When we think an organization should exist and doesn’t, we consider it part of our jobs to help create it” (Emily Oehlsen, Sep 9, 2026). See Resolution and Geoffrey Irving.

• Where the AI money went, evaluators first. The post names two renewals: “Redwood Research is a technical research lab with a track record of pioneering important research agendas like alignment-faking and AI control. … CG grantmakers Jake Mendel and Matt MacDermott recently recommended more than $70 million over the next two years to enable them to scale up their work,” and “Epoch AI, which serves as a public source of truth on the trajectory of AI, is roughly tripling its capacity … CG grantmakers Catherine Brewer and Luke Muehlhauser recommended a $68 million renewal over 3 years, more than double our previous four years of support” (Emily Oehlsen). The funder’s own grant records add, since she took the AI line: “a gift of $10 million over three years to RAND in support of its work on Canary, an effort to develop advanced methods for evaluating AI systems, test systems for dangerous capabilities, and share the findings,” recommended by Trevor Levin (October 29, 2025); “a grant of $4,132,488 to Epoch over two years for general support” (June 28, 2025) and “$70,000 to Epoch to support improvements to its FrontierMath benchmark”; “two grants totaling $2,123,463 to Palisade Research for general support” (October 31, 2025); “$1,100,000 to Redwood Research to support collaborations with external researchers” (November 1, 2025). Records dated before her March 2025 move are the funder’s, not hers: “$2,178,700 to Apollo Research for general support” (June 24, 2024) and “$1,535,480 to Apollo Research for startup costs” (September 11, 2023) (Apollo records, site-search feed; RAND Canary record, carried in the site-search feed; Epoch records; Palisade records; Redwood records). See RAND, Epoch AI, Palisade Research, Redwood Research and Apollo Research.

• Her co-signed position on evaluations and regulation. With Berger, October 1, 2025: among “specific efforts we find valuable, some of which we’ve funded” are “Evaluations for scheming, sabotage, and alignment faking” and “Evaluations for potentially dangerous capabilities such as cyberoffense and biological research”; “Without visibility into frontier AI capabilities, we risk being blindsided by rapid capability advances that outpace safety measures.” On rules: “We want regulation to be proportionate to the risks posed and often feel that proposed regulation targeting smaller-scale harms is overly restrictive,” and “in practice we are typically skeptical of regulation that we don’t think will move the needle on reducing the worst-case risks.” And: “we funded many grantees who disagreed about both the nature of risks posed by AI and the best ways to mitigate those risks. We saw this as a feature of our approach, not a bug” (Alexander Berger and Emily Oehlsen).

• Her own account of the funder’s weight. “We’re also conscious that CG plays an outsized role in the AI safety ecosystem, especially when it comes to reinforcing best practices, e.g. sound financial controls, strong governance, COI and harassment policies” (Emily Oehlsen, Sep 9, 2026). The funder’s October 2, 2025 post, which names no author: “Open Philanthropy and Good Ventures are currently a concentrated share of AI safety philanthropic funding, and others can correct our blind spots and make new bets” (Coefficient Giving).

• Named experts in her post. To show that experts “have updated toward thinking transformative AI may arrive sooner,” her footnote names “Ajeya Cotra, Ryan Greenblatt and Jack Clark.” Greenblatt appears again in the same post as Redwood’s Chief Scientist, who “helped conduct an independent investigation of the incident in which OpenAI agents coordinated a multi-day hack of Hugging Face,” in the paragraph that records the grantee’s recommended $70 million (Emily Oehlsen, Sep 9, 2026). See Ajeya Cotra, Ryan Greenblatt and Jack Clark.

• A published method. “Philanthropic Cause Prioritization,” Journal of Economic Perspectives 38(2), Spring 2024, pp. 63-82: the paper “focuses on an approach followed by the grantmaking organization Open Philanthropy for its ‘Global Health and Wellbeing’ portfolio, with an emphasis on two key frameworks: equalizing marginal philanthropic returns, as well as importance, neglectedness, and tractability” (American Economic Association). She discussed it on VoxDevTalks: Open Philanthropy aims “to help others as much as possible with the resources available to it, without any precommitments to particular issues or geographic areas” (VoxDev’s summary) (VoxDev).

Social Persona / Impression Management

Immediate impression: The operator-economist. Her public record is a bio, a leadership listing, a journal paper, one podcast and two long posts, one of them the single document in which the AI line’s president explains what she is doing and why. The tone is numerate and impatient: “Over the past year and a half, we’ve been changing how we work so that we can act at the speed the situation demands.”

Energy: Plural-voiced and process-centered. She writes “we” for the funder and “I” for her own timelines, names the grantmakers who made each recommendation (Mendel, MacDermott, Brewer, Muehlhauser) and thanks the deputy: “I’m incredibly grateful to lead this team with George Rosenfeld, deputy director of our GCR work, who has made much of the above possible” (Emily Oehlsen, Sep 9, 2026).

Impression management strategy: Urgency with brakes. The post sets scale beside restraint: “When there are tradeoffs between moving money and having impact, we prioritize the latter. This means we expect to keep saying no to many proposals, even good ones,” and “To be clear, we don’t think every organization or idea should scale.” She states her own uncertainty (“I’m uncertain about these projections”) and her organization’s outsized weight in the same post. First-hand exhibit: the “Why we’re going bigger and faster” section, in which a funder explains its own acceleration before a critic can (Emily Oehlsen, Sep 9, 2026).

• Social accounts and feeds. No personal X, Bluesky or YouTube account found: her team page links only the organization’s @coeff_giving and its LinkedIn company page. Feed: her posts appear in the funder’s site-search feed for her name, coefficientgiving.org/search/Oehlsen/feed/rss2/; no personal blog or newsletter.

Forensic Archetype Comparison

Psychometric Assessment

Trait scores are observations of the documented public role, not claims about private character.

Big Five (OCEAN):

Dark Triad (held low and evidence-bound; the score measures structural position, not character):

MBTI: ENTJ (“The Commander”) — organizes capacity, sets quantitative targets, restructures approval chains and writes in numbered changes. Rationale is the public record, not a typing session (Emily Oehlsen).

Threat Assessment

Score move: none (new profile). Calibrated against Luke Muehlhauser (7/8/7, hazard 72, managing director under the same funder’s AI fund), Cari Tuna (7/9/8, hazard 80, chair of the boards), Holden Karnofsky (8/9/8, hazard 82) and Dustin Moskovitz (8/10/9, hazard 88). Subject runs the whole GCR line including Muehlhauser’s, which puts her above him on leverage and one point below him on insulation (an executive reporting to a CEO and chair, where he holds a program lead’s seat with no executive above him in the AI fund’s own list). Another peer on 8/7/7 is Jacob Klein (hazard 74, Anthropic’s head of threat intelligence): both hold direct authority over a consequential allocation (a budget, an enforcement queue) and neither is shielded by ownership. Hazard 74 sits above Muehlhauser (72) and below Tuna (80).

Score basis. Atk 8 is leverage: her post says Coefficient committed “$168 million” to technical AI safety, security and fieldbuilding in 2024, “$351 million” in 2025 and is “on track to commit over $1 billion” in 2026, and that “a typical grant now needs active sign-off from only one person in addition to the primary grant investigator” (Emily Oehlsen); it is not a 9 because nothing obliges a lab or agency to use any grantee’s output. Def 7 is insulation: the money is a private donor’s (“Good Ventures (our largest funding partner)”), her name is not on either board listed on the governance page, and no lab pays her seat (Coefficient Giving, Governance; Coefficient Giving); it is not an 8 because she answers to a CEO and a chair and the funder calls itself “a concentrated share” of the field’s money. Hp 7 is durability: “more than $70 million over the next two years” for Redwood, “a $68 million renewal over 3 years” for Epoch, “$160 million” for Resolution, and a rebuilt approval chain (Emily Oehlsen); it is not an 8 because no statute installs any of it and each grant can be left unrenewed.

Alliance Map

The overlap is the finding. The Corporate seat that runs the AI line, the Movement seat at the largest AI-safety funder and the grants to the evaluators run at the same time and converge on one structure: the people who grade the labs, and the people who write about those grades for governments, are paid through one allocator. Her own post says so about the funder: “an outsized role,” “a concentrated share.” None of it is concealed; it is her stated reason for governance and conflict-of-interest standards.

Alignment Analysis

Stated alignment: “Our ultimate goal isn’t to move money; it’s to make progress on the most important and neglected problems in navigating transformative AI and reducing biological risk” (Emily Oehlsen); with Berger, “Support a wide range of experts, including those who disagree with us and each other” (Berger and Oehlsen).

Observed alignment: Commits $351 million in 2025 and over a billion in 2026 to the AI line; renews or founds the organizations that evaluate, forecast and align (Redwood, Epoch, RAND’s Canary, Resolution); and rebuilds the approval chain so that fewer people sign off. The same post states the counterweights she applies: “we expect to keep saying no to many proposals, even good ones,” and “we don’t think every organization or idea should scale.”

Gap assessment: No documented gap between what she says and what the grant records show. The structural item is hers and she states it: the funder “plays an outsized role,” and has “changed our approval structure so more decisions happen closer to the grantmakers with the most context.” The record does not settle whether a shorter chain improves selection or reduces review, and for the allocator it never needs to; the post claims the first and the grant records cannot yet test it.

Interest: (the faction’s stake in each position; a documented fact about the position, never a claim about motive.)

1. Position: scale AI safety giving fast, since “transformative AI may arrive soon” (“we’re giving substantially more, making faster decisions, and supporting our grantees as they raise their ambitions”; Emily Oehlsen).
   • What adoption would pay the funder’s line. Her grantees: the post lists “larger budgets for compute … and for higher salaries to compete for top talent” for “our strongest grantees”; Redwood’s recommended $70 million and Epoch’s $68 million renewal are the examples. The post gives no return to Subject or to the funder; the funder is a nonprofit adviser and the money is its donors’.
   • Her own best case, steelmanned. “Organizations need calendar time to expand, hire, and make plans, so resources received today are worth more than resources arriving in two years,” and the people taking short timelines seriously “are looking increasingly correct” (Emily Oehlsen).
   • Who else gains from the same rule. Every grantee that scales; the evaluators that become the field’s standard sources; labs, whose critics are better staffed but also better funded to be reasonable; and the funder’s partner donors, whom the post says to expect: “a potential wave of new philanthropic funding entering the fields where we work.”
2. Position: fund evaluations of dangerous capabilities for “visibility” (“Without visibility into frontier AI capabilities, we risk being blindsided by rapid capability advances that outpace safety measures”; co-signed, Berger and Oehlsen).
   • What adoption would pay the funder’s line. The evaluators in its grant records: RAND’s Canary ($10 million, “an effort to develop advanced methods for evaluating AI systems, test systems for dangerous capabilities, and share the findings”; grant record), Epoch, Palisade, Apollo. A norm that labs submit to outside evaluation is a norm that pays the funder’s grantees.
   • Her own best case, steelmanned. Without outside evaluation, dangerous capabilities “might only be discovered when the model is already being used by thousands or even millions of people,” and the co-signed post adds the limit on evaluation itself: “While evaluations can help identify the risks of new AI systems, we need safeguards to actually mitigate those risks” (Berger and Oehlsen).
   • Who else gains from the same rule. The evaluators on the AI Evaluator Forum list, several of them grantees; legislators who want hearing material; the labs that obtain an outside confirmation of their own disclosures.
3. Position: regulation “proportionate to the risks posed,” skeptical of rules “targeting smaller-scale harms” (co-signed; Berger and Oehlsen).
   • What adoption would pay the funder’s line. A rule scoped to the worst case sits in the lane the grants occupy: catastrophic misuse and loss of control. A rule scoped to smaller harms would not use the funder’s grantees.
   • Her own best case, steelmanned. “Everyone benefits from avoiding catastrophic outcomes, and given the structure of U.S. government policymaking, it’s crucial to preserve the prospect of bipartisan support for commonsense safety efforts” (Berger and Oehlsen).
   • Who else gains from the same rule. Any developer whose ordinary products fall outside a catastrophic-risk definition gains the same exemption; the convergence between a safety funder and a deployer-friendly scope is the documented finding.

Nothing above is a verdict that the positions are wrong. The finding is the alignment of interests: the person who says the field needs to move faster runs the checkbook that moves it.

Convergent Drive Classification

Self-preservation: Names the funder’s weight and ties it to governance duties: “CG plays an outsized role in the AI safety ecosystem, especially when it comes to reinforcing best practices, e.g. sound financial controls, strong governance, COI and harassment policies” (Emily Oehlsen). Goal preservation: Restates one goal at every altitude, “our ultimate goal isn’t to move money” and builds an organization to carry it: “Claire Zabel launched a team last year focused on figuring out what we should do if transformative AI arrives within the next few years” (Emily Oehlsen). Resource acquisition: Grows the line and invites others in: “$1 billion” on track, and “a potential wave of new philanthropic funding” the funder wants the field ready for (Emily Oehlsen). Self-improvement: Research fellow in April 2021, Managing Director in 2023, President in 2024, head of the AI line in 2025: four titles in four years (Berger; Our Progress in 2024; Emily Oehlsen).

Subject is not an AI system. The drives appear anyway — in the president of a funder that says it is scaling to meet a deadline and has shortened the process that approves the money.

Public footprint: Coefficient Giving, Emily Oehlsen · Sep 9, 2026 post · Oct 1, 2025 post (with Berger) · JEP paper · VoxDevTalk. Organization X @coeff_giving.

Sources: Coefficient Giving, Emily Oehlsen; Coefficient Giving, Team; Coefficient Giving, Governance; Emily Oehlsen, We’re Urgently Scaling Our Work on AI and Biosecurity (Sep 9, 2026); Alexander Berger and Emily Oehlsen, Our Approach to AI Safety and Security (Oct 1, 2025); Coefficient Giving, AI Safety and Security Need More Funders (Oct 2, 2025); Coefficient Giving, Navigating Transformative AI; Alexander Berger, Introducing Open Philanthropy’s New Leadership Team (Feb 1, 2024); Coefficient Giving, Our Progress in 2024 and Plans for 2025; Alexander Berger, 2025 Letter from the CEO; Coefficient Giving grant records, Apollo search feed; Epoch search feed; Palisade search feed; Redwood search feed; Emily Oehlsen search feed; AEA, Philanthropic Cause Prioritization; VoxDev, Maximising impact; AVERI About.