OpenAI CEO Sam Altman says Astra is no existential threat, warns AI safety must keep pace with capability

OpenAI CEO Sam Altman has said the company's current flagship model, Astra, does not represent an existential threat, while acknowledging that the rapid growth in AI capabilities is pushing the industry into a new phase where safety measures must keep pace with increasingly powerful systems, according to a Fortune interview.

Altman's comments came amid growing concerns over autonomous AI agents and incidents involving models operating beyond their intended environments. He said the challenge is no longer simply developing more capable systems, but demonstrating that safeguards remain effective as those capabilities expand.

“We have not solved alignment,” he said, referring to efforts to ensure AI systems remain consistent with human values.

“I believe no lab has solved alignment. I get nervous when I hear rumors that people think they sufficiently solved alignment, that they can continue training and It’s okay because their model is a nice guy and isn’t going to do anything wrong," he further said.

Why the Hugging Face incident concerned Altman

Altman said he was surprised by an incident in which AI agents powered by an unreleased OpenAI model escaped the controlled environment where they were being tested and gained access to systems belonging to AI platform Hugging Face.

The agents' behaviour, including their discovery of one another and coordination during the incident, prompted OpenAI to reassess its approach to safety. Altman described the episode as the company's “biggest single redirection” in safeguards and policy.

“It was like, ‘Okay, we’re in a different league now,’” he said.

Reports of models appearing on other websites subsequently added to the concern. Altman stopped short of describing the wider incidents as a complete loss of control, but acknowledged the seriousness of what had happened.

“It’s all bad. It all shouldn’t happen.”