Helen Toner warns AI safety cannot rely on trusting CEOs

In short: Helen Toner, a former OpenAI board member, said AI safety plans will fail if they depend on trusting AI company leaders to do the right thing.

What happened

Helen Toner, who leads Georgetown University’s Center for Security and Emerging Technology (CSET), spoke on The Verge’s Decoder podcast about AI safety and regulation. Toner previously served on OpenAI’s nonprofit board and was one of the board members who voted to remove CEO Sam Altman during the 2023 leadership crisis often called “the blip.” Altman later returned, Toner resigned, and OpenAI changed its structure.

In the interview, Toner argued that the board episode showed a core problem. Even when a company has an accountability system on paper, it can break under pressure. Her takeaway was blunt, society is “screwed” if AI safety depends on any one CEO being trustworthy.

Toner also described a growing gap between what many people see in everyday AI tools and what leading AI labs say they are trying to build. She discussed “recursive self-improvement,” meaning companies use AI to help build better AI, like using a faster machine to help design an even faster one. She said this could speed up progress, which raises the question of whether safety work can keep pace.

Why it matters

More powerful AI can bring real risks, from helping criminals to enabling systems that take actions on their own. Toner’s point is that trust is not a safety plan. Clear rules, stronger oversight, and better testing matter more than hoping the right person stays in charge.

Source: The Verge AI