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

Speaking to reporters after his DevDay keynote, Sam Altman said OpenAI will not go public until it can make confident safety claims about its models, with no firm timeline. He also said waiting too long would be “bad for the world.”

The comments land after the Hugging Face agent hack, further cybersecurity incidents across labs, Anthropic’s resignation-letter controversy, and industry talk of “pacing the frontier.” Anthropic has filed to go public with a reported November window; SpaceX (which owns xAI) already completed a landmark IPO in June.

We’re hearing a CEO try to hold two truths at once: safety first, but don’t wait forever for the public markets.

Altman’s cleaner line is the one about not wanting “additional pressure” from Wall Street while capabilities surge and safety requirements change. That is a real conflict. Public markets reward growth narratives; alignment work often looks like delayed launches. Saying he won’t “barrel, all guns blazing, towards an IPO” is the right instinct if the company means it.

I’m less impressed by vagueness on what “pacing” means. Pushing safety ahead of capabilities is a slogan until a specific model is held or a specific training run is stopped. OpenAI has paused training before; the question is whether that becomes the pattern or the press release.

Useful AI at scale needs capital. It also needs labs that can disappoint investors in the name of not shipping a misaligned agent into someone else’s infrastructure. If Anthropic IPOs first and OpenAI still cannot state its safety bar in falsifiable terms, the market will write the story for them.

We’re watching for the concrete claims Altman says he needs — not the IPO rumor mill.

The Verge, September 30, 2026 (Hayden Field), quoting Altman after DevDay. Prior context: Hugging Face hack disclosures, multi-lab incidents, Anthropic IPO filing, SpaceX IPO.