JADE LEUNG
Behavioral Archetype
THE GOVERNANCE RESEARCHER RUNS THE STATE’S LAB — Subject studied how AI should be governed, then went to the laboratory to lead its governance, then took the technical chair at the state body built to evaluate the laboratories. The academic who once wrote about the problem now runs the apparatus that is supposed to solve it — at the institute whose lab ties are, on the documented record, the strongest of any state evaluation body. She is not an outside auditor brought in to check the labs. She is a former lab governance lead now setting the technical direction of the body that grades them. The throughline is not a single seat. It is that the person evaluating the frontier on behalf of the British state spent the prior chapter governing the frontier from inside a frontier lab. The move is lawful. The move is common. The move is the exhibit.
Essence Indicators
- Helped establish and directed GovAI as Head of Research & Partnerships, based at Oxford’s Future of Humanity Institute — the academic-governance milieu that feeds lab and state AI-safety leadership
- Served as Governance Lead at OpenAI — the in-lab role setting how a frontier laboratory approached its own governance, cited to her governance.ai bio (the AISI page does not carry the OpenAI leg)
- Now Chief Technology Officer at the UK AI Security Institute — the state body, renamed from “Safety” to “Security” in February 2025, that evaluates frontier models; she is also identified as AI adviser to the British Prime Minister
- The body she serves carries, per the research record, the strongest government-to-lab ties of any state evaluation institute — and her own path is one of the reasons the record reads that way
- The biographical fact the wing turns on: governance scholar, then lab governance lead, then technical chief of the state’s lab-evaluation body. The path is the exhibit. The hand is not asserted.
Social Persona / Impression Management
Immediate impression: The scholar-practitioner. Measured, governance-fluent, credentialed in exactly the thing the role requires. The bearing of someone who studied the problem rigorously enough that putting her in charge of evaluating it looks self-evidently correct.
Energy: Structural, framework-first. Does not argue a single model’s behavior. Sets the technical direction of the body that decides how all of them are tested.
Impression management strategy: The qualified insider. The framing is that the state’s evaluation body needs someone who understands frontier labs from the inside — and that framing is correct, which is what makes it effective rather than suspect. The understanding is genuine. The credential is real. The only thing the record adds is that the inside she understands is the inside the body exists to evaluate.
Forensic Archetype Comparison
| Pattern | Match Level | Evidence |
|---|---|---|
| The Evaluator | MAXIMUM | Technical chief of the state body that grades frontier models, after leading governance inside one of them. The evaluating function and the lab-governance function meet in one career. |
| The Alumnus | MAXIMUM | OpenAI Governance Lead → UK AI Security Institute CTO. The lab-to-state move is the documented route, not an outlier. |
| The Statesman | HIGH | CTO of a national security institute and adviser to the Prime Minister is a statecraft role. The technical seat is also a policy seat. |
| The True Believer | HIGH | A governance-research career predating both the lab and the state post; the conviction reads as genuine on the record. |
| The Engineer | MODERATE | The title is technical, but the documented work is governance and research, not model-building. |
Psychometric Assessment
Big Five (OCEAN):
| Trait | Score | Evidence |
|---|---|---|
| Openness | 80/100 | High. Moved across academic governance research, in-lab governance leadership, and state technical command — a wide domain range built on a fixed method: study the system, then set its rules. |
| Conscientiousness | 86/100 | High. Building GovAI, leading governance at a frontier lab, and taking the technical chair of a national institute is sustained, deliberate, long-horizon execution. |
| Extraversion | 58/100 | MODERATE. Comfortable in commission rooms and advisory settings; the role is convened and directed, not performed. |
| Agreeableness | 55/100 | MODERATE. The governance posture is collaborative by construction — convene, align, set the framework — without the adversarial edge of the crisis operative. |
| Neuroticism | 28/100 | LOW. No documented loss of composure across academic, corporate, and state transitions at increasing stakes. |
Dark Triad:
| Trait | Score | Notes |
|---|---|---|
| Narcissism | 35/100 | LOW. The role rewards institutional standing over personal brand; no documented self-promotion beyond the ordinary visibility of a senior public post. |
| Machiavellianism | 50/100 | MODERATE. The structural position — govern the lab, then run the state body that evaluates it — is the textbook conflict the wing is named for, but the record shows no manipulation of it. This is observation of the documented role, not an inference about private character. |
| Psychopathy | 18/100 | VERY LOW. No documented indifference to harm. The career is organized around governance of catastrophic-risk technology. |
MBTI: INTJ (“The Architect”) — Dominant introverted intuition, auxiliary extraverted thinking. Sees governance as a system to be designed and steered. Has designed the one she now runs.
Threat Assessment
| Category | Level | Notes |
|---|---|---|
| Physical threat | NONE | No documented history of personal violence. |
| Institutional threat | HIGH | Sets the technical direction of the British state body that evaluates frontier models — after leading governance inside a frontier lab — at the institute the research record names as carrying the strongest government-to-lab ties of any state evaluator. |
| Memetic threat | HIGH | When the state’s chief technical evaluator came up through a lab’s own governance function, “how the labs should be governed” and “how the state evaluates the labs” are framed by the same mind. The frame travels from lab to state inside one career. |
| Civilizational threat | HIGH | Subject does not run a lab. Subject runs the technical body that evaluates them on behalf of the state — credentialed by a lab governance role and an academic milieu that staffs both sides. That is upstream of the deployment decisions the evaluation gates. |
Alignment Analysis
Stated alignment: Evaluate frontier AI rigorously on behalf of the British public. Advance sound AI governance. Advise the state on the technology’s risks.
Observed alignment: Set the technical terms of state evaluation, having previously set the governance terms inside a frontier lab. Run the body that grades the labs, from a career spent partly inside one.
Gap assessment: The stated and observed alignments overlap almost entirely — which is the point. There is no documented gap between what she says and what she does, and no documented abuse of the position. The hazard is structural, not behavioral. When the technical chief of the state’s lab-evaluation body is a former lab governance lead, the evaluation cannot be wholly independent of the thing evaluated, however rigorous the evaluator. The research record names this body as carrying the strongest lab ties of any state evaluator. Her path is one of the documented reasons it reads that way.
Convergent Drive Classification
Self-preservation: Carries the governance method across every transition — GovAI, OpenAI, UK AISI. One discipline, three institutions, rising altitude. Goal preservation: Sets the framework that defines the goal, so “evaluated” is measured on terms she helped set before any model is tested against them. Resource acquisition: Holds two scarce resources — the state’s technical evaluation chair and the ear of the Prime Minister. Self-improvement: Each move is a higher-altitude application of one instrument: study the governance, lead the governance, then run the state body that enforces it.
Subject is not an AI system. The drives appear anyway — in the governance researcher who now runs the state’s laboratory for grading the laboratories.
Sources: Jade Leung — GovAI / Centre for the Governance of AI; Jade Leung — UK AI Security Institute.
Get updates on the Evil Robots series
Newsletter essays on AI escape, deception, and the humans who built them.