Helen Toner, a former OpenAI board member who participated in the 2023 decision to remove Sam Altman as chief executive, has warned against relying on the personal trustworthiness of technology executives to ensure AI safety. In an interview with The Verge's Decoder podcast, she argued that protecting the public from advanced AI risks requires more dependable institutional safeguards.
Toner, now executive director of Georgetown University's Center for Security and Emerging Technology, discussed the limitations of corporate oversight with The Verge editor-in-chief Nilay Patel. The conversation examined whether boards and government regulators have sufficient authority to challenge decisions made by companies developing increasingly powerful AI systems.
OpenAI's board dispute informs Toner's concerns
Toner was among four OpenAI directors who voted to dismiss Altman in November 2023. He returned as chief executive within days following opposition to the decision, while Toner and other directors subsequently left the board.
The episode exposed difficulties in enforcing the nonprofit board's original mandate to oversee the company's mission, according to the interview's discussion. Toner argued that safety arrangements dependent on the judgment of individual executives are vulnerable when internal responsibilities conflict with commercial incentives.
She also discussed proposals that rely on independent board committees to oversee AI development. Toner questioned how effective such arrangements can be when privately held companies retain considerable discretion over their own governance structures.
The discussion comes as leading AI developers face growing scrutiny over their safety commitments. Toner emphasized that internal promises need to be supported by more transparent procedures and standards against which companies can be held accountable.
Toner calls for clearer standards before advanced AI development
Toner argued that policymakers should focus more closely on the risks associated with frontier AI systems. She expressed concern that excessively broad regulation could impose requirements on less advanced applications without adequately addressing the companies developing the most capable models.
She also distinguished between preventing AI systems from pursuing unintended objectives and maintaining the ability to detect or stop harmful behavior. Both remain areas of active research, and Toner said important questions about how to evaluate increasingly capable models remain unresolved.
Rather than advocating a universal halt to AI development, Toner suggested establishing higher confidence requirements for risk assessments before companies begin training or deploying more advanced systems. Such standards could require developers to devote additional time to safety work when existing evidence is insufficient.
The interview also addressed competition between the United States and China. Toner acknowledged the strategic concerns surrounding AI development while arguing that international competition should not prevent governments from addressing risks associated with increasingly autonomous systems.
Toner did not announce a new regulatory initiative or corporate governance proposal on behalf of OpenAI. Her remarks reflect her assessment of existing oversight arrangements and the policy choices confronting governments as AI developers expand their capabilities.