The OpenAI Board Member Who Fired Sam Altman Says We Still Can't Trust AI CEOs
CSET's Helen Toner argues that if our AI safety plan hinges on trusting any single person to get it right, "we're screwed."
In a wide-ranging conversation, Helen Toner—the former OpenAI board member who participated in the decision to remove Sam Altman as CEO in November 2023—reflects on what that episode revealed about the limits of trusting individual leaders in the AI industry. As of 2026, with the AI safety debate now entering its enforcement era in the U.S. and abroad, her warning carries fresh weight.
The Core Argument: No Single Person Should Be the Linchpin
Toner's central point is simple and uncomfortable: a governance model that depends on the good judgment or restraint of one charismatic CEO is structurally fragile. "We're screwed" if the plan to keep AI safe rests on any one person being trusted to get this right, she says. Instead, she argues for robust institutions—independent boards, external oversight, and regulatory frameworks—that do not collapse when an individual departs or exercises poor judgment.
What the 2023 Board Decision Signaled
The ouster of Sam Altman (and his swift reinstatement after employee and investor pressure) exposed how little leverage a board can have when commercial momentum and personal loyalty overpower formal governance. Toner's account underscores that the episode was not merely a corporate drama but a stress test for whether AI labs can hold themselves accountable—a test that, in her view, they largely failed.
2026 Context: From Principles to Enforcement
Since then, the stakes have only risen. By 2025–2026, major jurisdictions have shifted from voluntary AI safety commitments to binding rules—the EU AI Act's high-risk obligations are being phased in through 2026–2027, and U.S. federal agencies have begun formalizing model evaluation and reporting requirements. At the same time, frontier labs have grown more commercially powerful and more deeply embedded in critical infrastructure. Toner's argument maps directly onto this moment: as AI systems become more capable, the question of who holds the leash—and whether that person can be trusted—matters more, not less.
Takeaways for Policymakers and Boards
- Institutionalize oversight, don't personalize it. Safety should not depend on the character of a founder.
- Boards need real authority. Governance structures must be able to act independently of management and commercial pressure.
- External accountability is essential. Voluntary self-regulation has proven insufficient; regulation and third-party audits are needed to close the gap.
- Transparency is a precondition for trust. The public and regulators need visibility into how high-stakes decisions are made.
Toner's message remains a useful corrective to the industry's tendency to frame AI safety as a problem of individual virtue. In her view, the systems—legal, institutional, and organizational—must be designed to work even when the people in charge fall short.
This article is based on a podcast conversation with Helen Toner and has been edited for clarity. 2026 context added for publication.
via The Verge AI