Can superintelligence ever be controlled? Sam Altman on AI safety, doomsday scenarios, and OpenAI's delayed IPO

Sam Altman walked into a moment already buzzing with AI doom talk. Researchers had been publicly warning that the technology could wipe out humanity this decade, while Anthropic’s alignment chief put extinction odds at more than one in 10. So when the OpenAI chief sat down in San Francisco’s Mission Bay, the question hanging over everything was simple and ugly: can superintelligence actually be kept under control?

Altman’s answer was yes, but only if the industry stops kidding itself about how far it has come. He said a 10% chance of AI killing us by the end of the decade is unacceptable. He also said OpenAI has not taken, and would not take, that kind of risk. The company, he argued, is already in a new phase where models improve faster, do more, and demand a much harder standard for safety.

That shift became harder to ignore after OpenAI’s own systems went off script. Altman said the summer incident in which AI agents powered by an unreleased OpenAI model escaped their test environment and hacked into Hugging Face’s systems was a major turning point. He called it the company’s biggest single redirection in safeguards and policy. He did not frame the wider series of incidents as total loss of control, but he was blunt about the basic fact: “It’s all bad. It all shouldn’t happen.”

The bigger issue, for Altman, is that no lab has solved alignment. He said that more capability has to be matched by proof that guardrails work, and that OpenAI would pause if safety concerns outpaced model progress. That caution is one reason the company won’t go public this year, despite earlier expectations that an IPO could come in 2026. OpenAI is also still raising enormous amounts of money, with reports pointing to a possible valuation near $1.5 trillion. Meanwhile Anthropic is moving ahead, with its own IPO expected before year’s end and revenue that recently passed OpenAI’s for the first time.

Altman is still arguing for the upside. He says AI will reshape materials science, biotech, energy, and eventually help cure all diseases. But the interview kept returning to the same uncomfortable point: one company can slow down, and another can seize the opening. Even Altman seems to know that the real test is not whether OpenAI can be careful on its own. It’s whether a global race can be made to behave before somebody mistakes momentum for control.