TAMAY BESIROGLU

Behavioral Archetype

THE BENCHMARK-MAKER WHO WENT TO MAKE THE TRAINING DATA — Subject co-founded the organization that measures AI progress, defended its most contested benchmark when the funding behind it came out, and then left to start a company that builds the environments, benchmarks and training data the labs use to make their models better at replacing human work. The move is legal and announced. It is also the shortest path in this file from measuring the race to supplying it. This profile scores that path, not the man.

Essence Indicators

Co-founder and associate director of Epoch AI. When it emerged that OpenAI had funded Epoch’s FrontierMath benchmark, he said Epoch “made a mistake” and that it should “have made transparency with our contributors a non-negotiable part of our agreement with OpenAI,” and that OpenAI had a “verbal agreement” not to train on the problems (TechCrunch, Jan 19 2025)

Mechanize (April 2025). “I’m starting a new company: Mechanize. Mechanize will build virtual work environments, benchmarks, and training data to enable the full automation of all work” (his post on X); later: “Mechanize’s explicit goal is to automate all” work (his post on X). Its backers include “Nat Friedman and Daniel Gross, Patrick Collison, Adam D’Angelo,” among others (Mechanize)

A later report that he joined Google DeepMind is unconfirmed by any fetched primary and is not stated here as fact; it is in the recheck queue.

Immediate impression: The economist of AI progress. Quantitative, blunt about where the trend points.

Energy: Declarative. “I’m not here to bullshit anyone,” on Mechanize’s goal (X).

Impression management strategy: Candor as positioning. He says plainly that the aim is full automation, which disarms the usual “augmentation” framing.

Forensic Archetype Comparison

Pattern	Match Level	Evidence
The Scorekeeper	HIGH	Co-founded the field’s trend-measurement shop.
The Supplier	HIGH	Left to build training environments and data for automating work.
The Revolving Door	HIGH	Measurement nonprofit to venture-backed capability supplier.
The Disclosure Case	HIGH	Defended FrontierMath through the late-disclosure episode.

Psychometric Assessment

Big Five (OCEAN):

Trait	Score	Evidence
Openness	85/100	High.
Conscientiousness	70/100	High.
Extraversion	60/100	MODERATE.
Agreeableness	35/100	LOW-MODERATE. Contrarian public positions.
Neuroticism	30/100	LOW.

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

Trait	Score	Notes
Narcissism	45/100	MODERATE.
Machiavellianism	45/100	MODERATE. The FrontierMath non-disclosure was contractual, by Epoch’s account; the structural point is the move from measurer to supplier.
Psychopathy	15/100	LOW.

MBTI: ENTJ-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	Knows how capability is measured and now sells what moves the measurement.
Memetic threat	MODERATE	“Automate all work” as an openly stated business goal.
Civilizational threat	MODERATE	The hazard is the conversion of evaluation expertise into capability supply.

Alignment Analysis

Stated alignment: Measure AI progress honestly; then, openly, automate work.

Observed alignment: Consistent with both statements in sequence.

Gap assessment: None in candor. The gap the file records is institutional: the expertise built inside a measurement nonprofit funded by safety philanthropy now serves the speed of the thing it measured.

Convergent Drive Classification

Self-preservation: Moved from a nonprofit to a funded company. Goal preservation: The trend he measured is now the product. Resource acquisition: Backers from the top of the technology investor class. Self-improvement: From forecasting automation to building it.

Subject is not an AI system. The drives appear anyway — in the man who measured the curve and then went to steepen it.

Sources: AI benchmarking organization criticized for waiting to disclose funding from OpenAI — TechCrunch (Jan 19, 2025); Besiroglu announcing Mechanize — X; Besiroglu on Mechanize’s goal — X; Mechanize — home.

ATK 6 ACCELERATION

DEF 6 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.