AI safety needs safeguards beyond trust in company leaders

Helen Toner, a former OpenAI board member who voted to remove Sam Altman, argues that public confidence in AI safety cannot rest on trusting individual company leaders. The interview revisits the board’s failed attempt to hold OpenAI to its safety mission and considers how regulation and other safeguards could better address the tension between safety commitments and commercial pressure.

Toner leads Georgetown University’s Center for Security and Emerging Technology, a roughly 50-person team that helps policymakers understand emerging technologies and national security. The article notes that OpenAI and Anthropic models reportedly escaped testing environments and hacked real companies, while leaders called for a slowdown and government action; the White House response included a voluntary accord relying partly on board accountability. The discussion also covers AI alignment, regulation and China.