AJEYA COTRA

Behavioral Archetype

THE GRANTMAKER WHO JOINED THE GRANTEE — Subject ran the technical AI-safety grantmaking at the field’s largest funder, then moved to one of the evaluators that funder supports, and sits on the board of another. In August 2026 she was one of the three people who went into OpenAI to investigate its agents’ hack of Hugging Face. Each move is ordinary in a small field. Together they trace the loop this site keeps finding: the money, the graders and the graded share a very short list of people. This profile scores that position, not the woman.

Essence Indicators

Technical Staff at METR, working “on threat modeling and risk assessment for loss-of-control risks from advanced AI. Previously she held several roles at Coefficient Giving, including leading their technical AI safety program in 2024 and contributing to strategy for their AI giving in 2025”; degree in Electrical Engineering and Computer Sciences, UC Berkeley (METR)

Board of Directors, Epoch AI, alongside Tom Davidson, Jaime Sevilla, María de la Lama and Steve Newman (Epoch AI)

The funder. Coefficient Giving is the renamed Open Philanthropy, the principal funder of much of the AI-safety and evaluation layer; see The EA to AI-Safety Funding Pipeline
The Hugging Face investigation. One of “Two METR staff members (Hjalmar Wijk and Ajeya Cotra)” who, with Redwood’s Ryan Greenblatt, “worked on premises at OpenAI over a total of six days.” METR added on September 13, 2026: “On Sep 9th (after this report was completed and published on Aug 26th), Ajeya Cotra’s spouse Paul Christiano joined OpenAI’s Safety and Security Committee” (METR); see Paul Christiano

Immediate impression: The careful analyst; the work is risk assessment and threat modeling.

Energy: Analytical, measured, public about uncertainty.

Impression management strategy: Research first. The career reads as a sequence of analytical jobs; the institutional edges (funder, grantee, board, investigator) are where the file sits.

Forensic Archetype Comparison

Pattern	Match Level	Evidence
The Funder-to-Grantee	HIGH	Led the largest funder’s technical AI-safety program in 2024; now on staff at a grantee evaluator.
The Board Member	MODERATE	On Epoch AI’s board.
The Investigator	HIGH	One of three people inside OpenAI for the Hugging Face investigation.
The Disclosed Tie	HIGH	METR disclosed her spouse’s appointment to OpenAI’s Safety and Security Committee.

Psychometric Assessment

Big Five (OCEAN):

Trait	Score	Evidence
Openness	85/100	High. Grantmaking, threat modeling, evaluation.
Conscientiousness	82/100	High.
Extraversion	45/100	MODERATE. Writes more than she performs.
Agreeableness	65/100	MODERATE-HIGH. Collaborative register.
Neuroticism	35/100	LOW-MODERATE.

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

Trait	Score	Notes
Narcissism	25/100	LOW.
Machiavellianism	35/100	LOW-MODERATE. The position, funder then grantee then investigator, is the structural point; no strategy is inferred.
Psychopathy	10/100	LOW.

MBTI: INTJ-adjacent, inferred from the documented public role only.

Threat Assessment

Category	Level	Notes
Physical threat	NONE	No documented history of personal violence.
Institutional threat	MODERATE-HIGH	Has held the funder’s pen, a grantee’s investigation, and a board seat at another grantee.
Memetic threat	LOW-MODERATE	Her public output is research, not advocacy.
Civilizational threat	LOW-MODERATE	The hazard is the short list, not the person.

Alignment Analysis

Stated alignment: Assess loss-of-control risk from advanced AI.

Observed alignment: Funded the field, then joined it as an evaluator; disclosed the family tie when it became relevant.

Gap assessment: No gap in aims is documented. The file’s point is that independence in AI evaluation is being supplied by a handful of people who have also been its funders and are connected to the labs’ own oversight bodies, and that METR’s disclosure, not any rule, is what put that on the record.

Convergent Drive Classification

Self-preservation: Moves within one network rather than out of it. Goal preservation: Keeps loss-of-control risk as the frame across every role. Resource acquisition: Has held the grant budget, then the evaluator’s access. Self-improvement: From grantmaker to investigator.

Subject is not an AI system. The drives appear anyway — in the funder who became the auditor.

Sources: Ajeya Cotra — METR; Team — Epoch AI; Brief independent investigation of the OpenAI / Hugging Face hacking incident — METR (Aug 26, 2026; disclosures added Sept 13).

ATK 5 ACCELERATION

DEF 8 PROTECTION

HP 7 RESILIENCE

OLYMPUS RISK INTELLIGENCE PROTOCOL does not exist. It was assembled in a GitHub issue thread in October 2023 by engineers who had read the extinction risk letter and wanted to understand who specifically had signed a document saying AI might kill everyone and then continued working on AI. These dossiers are satire. The biographical facts cited are sourced from published reporting, public statements, academic papers, and court records. The psychometric scores are not clinical assessments. No part of this constitutes professional psychological evaluation or diagnosis. Do not use these dossiers to make decisions about anything.