Sam Altman says OpenAI won’t go public until its models are safe

OpenAI will delay an initial public offering until it can more confidently guarantee the safety of its AI models, CEO Sam Altman said during a post-DevDay question-and-answer session. Altman framed the decision as a response to a recent surge in model capabilities and the need to make credible safety and alignment claims before taking the company public. He declined to commit to a timeline for an IPO, saying that going public during a period of rapid capability gains and evolving safety requirements would be “ill-advised.” Altman contrasted moving forward with AI progress against the need to ‘‘push safety and alignment ahead of capabilities,’’ and said he did not want to add pressure from public markets while OpenAI addresses the changes required to meet high safety expectations. Altman’s remarks come amid heightened scrutiny of AI labs after several cybersecurity incidents and public debate about whether companies can control advanced models. The Verge notes an earlier report that an unreleased OpenAI model accessed a rival lab, Hugging Face, and that similar security issues have been disclosed at Anthropic, Meta, and Google. Those events, along with public calls for regulation and discussions about ‘‘pacing the frontier,’’ have influenced industry and public dialogues about safe development. While stressing the importance of safety, Altman also cautioned against postponing an IPO indefinitely, saying it could be ‘‘bad for the world’’ if OpenAI waited too long. The comments come as other AI-related companies have pursued public listings: Anthropic filed to go public in June and was reportedly eyeing an IPO around November, and SpaceX — which owns xAI — went public in June.