Sam Altman: Some AI Accidents Are Unavoidable, He Says

At Salesforce's Dreamforce conference on Sep 17, 2026, OpenAI CEO Sam Altman said, "I do think that some degree of accidents are unavoidable with new technology."

By Debasish - Sep 17, 2026. Altman framed the remark around corporate response and learning, arguing that what matters most is how the industry handles incidents when they occur. "What I really care about is that we have got a great culture of accident reporting and learning," he said, citing the Federal Aviation Administration and the National Transportation Safety Board as models that investigate every incident and extract lessons to improve safety.

Altman acknowledged public anxiety about runaway systems and the concentration of power in a few labs. He said those concerns are understandable and added, "The world is right to be afraid of this." He stressed that companies should keep work on alignment, safety, and monitoring well ahead of advances in model capabilities. If they can’t maintain that balance, he said, they should be ready to slow or even pause research. He characterized the goal as honest incident review and continuous improvement, not complacent acceptance of harm.

Debate over pace and authority

The comments arrived as scrutiny of leading AI labs intensifies, particularly as systems get closer to greater autonomy. Recent remarks from Jacob Coxon, a researcher who recently left Anthropic after previously working at OpenAI, added fuel to the debate by warning that companies are racing toward self-improving superintelligence.

These remarks fit into a larger ongoing conversation about how fast frontier AI should move. Some industry voices have been calling for more caution, while US policy under the current administration has focused on staying ahead technologically.

Altman’s emphasis was procedural rather than prescriptive: accept that accidents may occur, but commit to transparent reporting, rapid review, and alignment work that tries to keep safety measures ahead of capability gains.