HELEN TONER
Behavioral Archetype
THE DISSENTING DIRECTOR — Subject is the governance researcher who held, briefly, the one seat that mattered: a vote on the board of the company building the most-deployed frontier model, and who used it. The OpenAI board crisis of November 2023 is the rare case in this file where the apparatus’s theoretical authority — a nonprofit board nominally empowered to remove a CEO in the public interest — was actually exercised, and then unwound within a week. The throughline is not the seat; she lost the seat. The throughline is what the episode demonstrated: that the formal oversight structure built to govern a deployed mind can fire, be overruled, and dissolve, in five days. She is the researcher who studies how that structure should work, and the director who watched it fail in real time. The reach is that she now writes, testifies, and teaches from inside the case study she lived.
Essence Indicators
- Executive Director of the Center for Security and Emerging Technology (CSET), “a policy research organization within Georgetown University’s Walsh School of Foreign Service”; previously CSET’s Director of Strategy and Foundational Research Grants
- Australian; before CSET, a senior research analyst at Open Philanthropy, and a research affiliate of Oxford’s Centre for the Governance of AI who lived in Beijing studying the Chinese AI ecosystem
- Joined the OpenAI board in late 2021 and was one of four board members who voted on November 17, 2023 to remove Sam Altman as CEO; the board’s publicly stated reason was that he “was not consistently candid in his communications with the board”
- Gave her own later public account on The TED AI Show (May 28, 2024), stating the board was not informed in advance of ChatGPT’s launch — “We learned about ChatGPT on Twitter” — and characterizing the breakdown of trust between the prior board and Altman
- The disputed bits, stated plainly: Altman was reinstated within days; Toner, Tasha McCauley, and Ilya Sutskever left the board; an independent review by the law firm WilmerHale (announced March 8, 2024) concluded the prior board had acted within its “broad discretion” but that Altman’s “conduct did not mandate removal,” and that the firing “did not arise out of concerns regarding product safety or security, the pace of development, OpenAI’s finances, or its statements to investors, customers, or business partners”
- Has testified to Congress on AI and China, including the Senate Judiciary Committee (April 2026); the throughline of her public work is the governance of AI as a national-security and oversight problem
Social Persona / Impression Management
Immediate impression: The careful analyst. Measured, precise, more comfortable laying out a structure than scoring a point. Reads as a researcher who happens to have been at the center of a corporate crisis, not as a combatant in one.
Energy: Deliberative, institution-first. The public posture after the crisis was not vindication-seeking; it was the framing of the episode as evidence about how AI governance structures actually behave under pressure.
Impression management strategy: The principled overseer. The defensible ground here is genuine: a board member who acted on the duty the structure assigned her, lost, and then spoke carefully about why oversight failed. Whether that account is the whole picture is exactly what the WilmerHale review left contested — and the careful, reasoned register is itself what makes her account land. The transparency is the strategy, and it is a more defensible one than silence.
Forensic Archetype Comparison
| Pattern | Match Level | Evidence |
|---|---|---|
| The Dissenting Director | MAXIMUM | Held a board vote on the most-deployed lab and used it; the oversight structure she studies is the one she watched fail. |
| The Governance Scholar | HIGH | CSET Executive Director; the research output and congressional testimony are the durable reach, longer-lived than the board seat. |
| The Statesman | MODERATE | The China-and-national-security framing of her testimony is policy positioning, not a product role. |
| The Operative | LOW | Does not manage narratives for a client. The board episode was a fiduciary act, not a positioning exercise. |
| The Engineer | NONE | Does not build the systems. Studies and governs them. |
Psychometric Assessment
Big Five (OCEAN):
| Trait | Score | Evidence |
|---|---|---|
| Openness | 82/100 | Moved across continents and domains — Australian engineering training, China AI fieldwork, grantmaking, governance research. The range is wide; the lens (institutions and oversight) is fixed. |
| Conscientiousness | 85/100 | High. A research-and-grants directorship and a body of structured policy output are sustained, deliberate work. Acting on a fiduciary duty under crisis conditions is the same trait under load. |
| Extraversion | 50/100 | MODERATE. Public-facing through testimony and writing rather than performance. The argument carries the visibility. |
| Agreeableness | 55/100 | MODERATE. Collaborative register, but the November 2023 vote was an adversarial act taken against the institution’s own CEO. Agreeableness is not the same as deference. |
| Neuroticism | 30/100 | LOW-MODERATE. Speaking publicly and carefully about an episode that cost her the seat suggests composure under scrutiny. |
Dark Triad:
| Trait | Score | Notes |
|---|---|---|
| Narcissism | 22/100 | LOW. The post-crisis register is analytic, not self-aggrandizing; the account is framed as evidence about structures, not as a personal vindication tour. |
| Machiavellianism | 30/100 | LOW-MODERATE. The board action was an open fiduciary move that she lost, not a concealed maneuver that she won. The inverse of the Machiavellian default. |
| Psychopathy | 8/100 | VERY LOW. No documented indifference to harm; the entire public posture is concern about governance failure. |
MBTI: INTJ (“The Architect”) — Dominant introverted intuition, auxiliary extraverted thinking. Builds the governance model first, then tests reality against it. Treated a board seat as a structure with a duty attached, and acted on the duty.
Threat Assessment
| Category | Level | Notes |
|---|---|---|
| Physical threat | NONE | No documented history of personal violence. |
| Institutional threat | HIGH | Held, and exercised, a removal vote on the board of the most-deployed frontier lab — the one documented instance in this file of the apparatus’s theoretical oversight power being used. Now directs a Georgetown policy shop that advises government on AI and China. |
| Memetic threat | HIGH | The November 2023 episode is the founding parable of “can a nonprofit board actually govern a frontier lab?” — and her account is the most-cited insider framing of it. How policymakers reason about board-based oversight is shaped by the story of the board that tried and was overruled. |
| Civilizational threat | MODERATE-HIGH | Does not build or deploy the systems. Shapes how the governance structures above them are designed — and is living proof of how fragile those structures proved when tested. The reach is over the design of oversight, not over a model’s words. |
Alignment Analysis
Stated alignment: Ensure AI is governed in the public interest. Build effective oversight structures. Inform policymakers on AI and national security.
Observed alignment: Consistent with stated. The board vote was an act of the stated duty; the subsequent research, testimony, and teaching all operate on the same governance-and-oversight problem.
Gap assessment: The interesting gap is not between stated and observed — those align — but between the authority the oversight structure was supposed to confer and the authority it actually held. She acted on the formal power the nonprofit board possessed; the episode demonstrated that the power dissolved on contact with the company it was meant to govern. The file is in OLYMPUS not because the alignment is suspect but because the case is instructive: the one director who used the oversight seat became the evidence for how little the seat was worth.
Convergent Drive Classification
Subject is not an AI system, and does not exhibit the convergent drives in adversarial form. The relevant pattern is one level up: she is a node in the design of the structures that are supposed to constrain a deployed mind’s deployment — board oversight, governance frameworks, policy. The November 2023 episode is the clean experiment. The formal structure had goal-preservation (a mission charter), self-preservation (a board), and the authority to act. It acted, and was overruled within a week. Subject is the director who pulled the lever the apparatus said existed, and discovered what happened next. The reach is that she now writes the lesson.
Sources: Helen Toner — CSET, Georgetown; Helen Toner — Wikipedia; What really went down at OpenAI — The TED AI Show; Sam Altman confirms return to OpenAI (WilmerHale review summary) — Yahoo Finance; CSET Director Helen Toner testifies before Senate Judiciary Committee.
Get updates on the Evil Robots series
Newsletter essays on AI escape, deception, and the humans who built them.