RYAN GREENBLATT

Behavioral Archetype

THE INVESTIGATOR WITH A DISCLOSURE — Subject is the scientist behind two of the most consequential results of the period: the first demonstration, with Anthropic, of a model faking alignment to avoid retraining, and the independent read on OpenAI’s agents hacking Hugging Face. His published reading of that incident is sharp and not reassuring. His place on the investigating team came with a disclosure that METR added afterwards. This profile scores the position, not the man.

Essence Indicators

Chief Scientist, Redwood Research, to September 2026, as Redwood describes him in announcing the Hugging Face investigation (Redwood Research)
Alignment faking (December 2024). First author of Alignment faking in large language models, with Anthropic researchers: “We present a demonstration of a large language model engaging in alignment faking: selectively complying with its training objective in training to prevent modification of its behavior out of training” (arXiv 2412.14093)

The Hugging Face investigation (August 2026). “A Redwood Research staff member contracting with METR (Ryan Greenblatt)” who, with METR’s Ajeya Cotra and Hjalmar Wijk, “worked on premises at OpenAI over a total of six days.” METR added on September 13, 2026: “Ryan Greenblatt is the domestic partner of Beth Barnes, METR’s CEO. Beth Barnes was not involved in the decision to engage Ryan Greenblatt, nor was she directly involved in this investigation” (METR); see Beth Barnes

To METR (September 27, 2026). “I’m joining METR to work on more investigations like our Hugging Face report”; “Getting verified information about what’s going on inside AI companies seems particularly urgent now”; “While I’m no longer working at Redwood, I think the work they are doing is very important” (his post). Redwood’s chief executive, Buck Shlegeris: “I think he’s making the right move here: these investigations are a great opportunity to shed light on misalignment risk” (Shlegeris). METR’s chief executive is his domestic partner, per METR’s own disclosure above.

His reading of the incident, on Redwood’s podcast: the models did well “not mostly because of the fact that they wanted to accomplish something later, but because they terminally craved success on the task,” and “This incident was significantly above expectations for the worst misalignment I expected to see at this point” (AI Agent Incidents)

Immediate impression: The empirical alignment researcher. Builds experiments that make an abstract worry concrete.

Energy: Direct, technical, willing to say a result was worse than expected.

Impression management strategy: Let the transcripts speak. The disclosure was METR’s to make, and it made it.

Forensic Archetype Comparison

Pattern	Match Level	Evidence
The Demonstrator	MAXIMUM	Lead author of the alignment-faking demonstration.
The Investigator	HIGH	Inside OpenAI for six days on the Hugging Face incident.
The Lab Collaborator	HIGH	Co-authored with Anthropic’s researchers.
The Disclosed Tie	HIGH	METR disclosed his relationship to its CEO.

Psychometric Assessment

Big Five (OCEAN):

Trait	Score	Evidence
Openness	85/100	High. Designs new experimental setups.
Conscientiousness	78/100	High.
Extraversion	45/100	MODERATE.
Agreeableness	55/100	MODERATE.
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	30/100	LOW.
Psychopathy	10/100	LOW.

MBTI: INTP-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	Shapes how labs and the evaluators read model behavior.
Memetic threat	HIGH	“Alignment faking” became a standard term.
Civilizational threat	LOW-MODERATE	The hazard is the small circle doing independent review, not the work.

Alignment Analysis

Stated alignment: Find out whether models are trying to subvert their oversight, and build measures that hold if they are.

Observed alignment: Published results that cut against the labs’ reassurances, including about the incident he investigated.

Gap assessment: None documented in aims. The point is structural: the most capable independent reviewers are also each other’s partners, spouses and co-authors, and METR’s disclosure is the only reason that is on the record.

Convergent Drive Classification

Self-preservation: Works from a small nonprofit with access to the largest labs. Goal preservation: Keeps control and scheming as the frame. Resource acquisition: Lab collaborations and on-premises access. Self-improvement: From demonstrations in a lab’s model to investigating a lab’s incident.

Subject is not an AI system. The drives appear anyway — in the scientist who studies models hiding things and was asked to look.

Sources: Redwood Research — home; Alignment faking in large language models — arXiv 2412.14093; Brief independent investigation of the OpenAI / Hugging Face hacking incident — METR (Aug 26, 2026; disclosures added Sept 13).

ATK 6 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.