Helen Toner: AI Company Leadership Cannot Be Taken at Its Word
Helen Toner was one of four OpenAI board members who voted to remove Sam Altman as CEO in November 2023. She has since spoken more openly about what drove that decision — and her broader conclusions about the governance of AI companies extend well beyond OpenAI specifically. Her position is direct: the structural conditions that allowed the board's concerns to be overridden have not been corrected, and the industry has not produced meaningful accountability mechanisms in the time since.
Toner's criticism is not primarily about individual executives. It is about the absence of institutional frameworks capable of constraining them. In her assessment, the current operating environment allows AI company leaders to make consequential decisions — about deployment timelines, safety thresholds, and product scope — without sufficient external oversight or internal checks that carry real authority.
The OpenAI episode illustrated the problem in concrete terms. A board with nominal governance authority attempted to act on safety-related concerns and was effectively reversed within days, following pressure from investors, employees, and partners. The structure that was supposed to provide accountability dissolved under commercial pressure. Toner's argument is that this failure mode is not unique to OpenAI — it reflects a sector-wide governance gap.
What Toner is pointing toward is a specific category of institutional risk: organizations that communicate a commitment to safety as a cultural and operational priority, while maintaining decision-making structures that cannot enforce that commitment when it conflicts with business interests. The gap between stated values and enforceable governance is, in her framing, the core problem.
For companies integrating AI systems into their operations, this analysis carries practical weight. Vendor relationships with AI developers involve an implicit trust in the internal decision-making of those developers — about when models are ready for deployment, what risks have been evaluated, and what constraints govern capability releases. If those internal processes are not structurally sound, downstream enterprise users absorb risk they cannot observe or control.
Regulatory frameworks have moved slowly relative to the pace of deployment. The EU AI Act provides some structural requirements, but enforcement timelines are long and technical compliance does not resolve questions of organizational integrity. In the United States, binding sector-specific AI governance remains largely absent. Toner has been involved in policy work through Georgetown's Center for Security and Emerging Technology, and her public positioning suggests she views the policy gap as a durable problem, not a transitional one.
The longer-term signal here is that the credibility of AI developers as institutions — not just the capability of their models — will increasingly factor into how enterprises, regulators, and governments structure their AI dependencies. A period of rapid deployment followed by high-profile governance failures could produce a tightening of oversight requirements that companies are not currently anticipating. The OpenAI board episode was treated by many observers as an internal corporate dispute. Toner's framing positions it as an early data point in a broader pattern that the industry has not yet taken seriously.
The question is not whether AI executives are acting in bad faith. It is whether the structures around them are capable of producing consistent, verifiable behavior aligned with the commitments those executives make publicly. On that question, Toner's answer remains no.
Sources: — The Verge (https://www.theverge.com/podcast/1032907/helen-toner-openai-cset-ai-safety-policy-interview)