OLYMPUS RISK INTELLIGENCE PROTOCOL — HUMAN THREAT ASSESSMENT DIVISION CASE WTW-2026-158

ADAM GLEAVE

THE APPARATUS THE MAN WHO BEAT THE SUPERHUMAN GO ENGINE
EVALUATION WING — THE RED-TEAMER ON THE OTHER EVALUATOR'S BOARD
Status
ACTIVE — co-founder and CEO, FAR.AI; board member, METR, Safe AI Forum and London Initiative for Safe AI
Hazard
55
ATK / DEF / HP
6 / 7 / 6

Behavioral Archetype

THE MAN WHO BEAT THE SUPERHUMAN GO ENGINE — Subject co-led the research showing that a superhuman Go program could be beaten by a simple adversary, then built the nonprofit that does pre-deployment testing for frontier developers and runs the field’s workshop circuit. He sits on the board of METR, the other leading evaluator. This profile scores the position, not the man.

Essence Indicators

  • “Founder & CEO at FAR.AI.” He sits on the boards of the Safe AI Forum, the London Initiative for Safe AI and METR; previously at Google DeepMind with Jan Leike and Geoffrey Irving; PhD at UC Berkeley under Stuart Russell (Adam Gleave). See FAR.AI.
  • The Go result. Co-first author of Adversarial Policies Beat Superhuman Go AIs (ICML 2023): the adversary won more than 97 percent of games against KataGo at superhuman settings (ICML 2023).

Forensic Archetype Comparison

PatternMatch LevelEvidence
The Interlocking BoardHIGHRuns one evaluator, sits on another’s board.
The Lab AlumnusMODERATEGoogle DeepMind.
The ConvenerHIGHFAR.AI’s Alignment Workshop series.

Psychometric Assessment

Big Five (OCEAN): Openness 80, Conscientiousness 75, Extraversion 55, Agreeableness 60, Neuroticism 35 (inferred from the public role only).

Dark Triad (evidence-bound): Narcissism 30, Machiavellianism 30, Psychopathy 5.

Threat Assessment

CategoryLevelNotes
Physical threatNONENo documented history of personal violence.
Institutional threatMODERATEHelps govern two of the field’s evaluators.
Memetic threatLOWA technical voice.
Civilizational threatLOWStructural, not personal.

Alignment Analysis

Stated alignment: Understand and reduce AI risk. Observed alignment: Consistent. Gap assessment: None in aims. The point is the interlock: the evaluators check the labs, and the same few people govern the evaluators.

Convergent Drive Classification

Self-preservation: Runs his own nonprofit. Goal preservation: Robustness and security research. Resource acquisition: Philanthropic funding and board seats. Self-improvement: From a PhD result to a seat on the field’s governing boards.

Subject is not an AI system. The drives appear anyway.

Sources: Adam Gleave — personal site; About — FAR.AI; Adversarial Policies Beat Superhuman Go AIs — ICML 2023.

ATK 6 ACCELERATION
DEF 7 PROTECTION
HP 6 RESILIENCE

Cross-Referenced Files — THE APPARATUS

#148 58 JEFFREY LADISH THE SECURITY MAN WHO LEFT TO WARN #156 56 BUCK SHLEGERIS THE MAN WHO ASSUMES THE MODEL IS AGAINST YOU #154 55 GEOFFREY IRVING THE LAB SCIENTIST WHO RAN THE GOVERNMENT'S SCIENCE #160 50 JACOB STEINHARDT THE AUDITOR WHO NEEDS THE KEYS
ALL THE APPARATUS FILES →
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.