Altman embraces inevitable accidents

At Salesforce's Dreamforce conference, OpenAI's CEO said some AI accidents are unavoidable with new technology. What matters is having a strong culture of transparent accident reporting and learning from incidents.

“I do think that some degree of accidents are unavoidable with new technology.”

Aviation industry as the model

Altman cited the FAA and NTSB as examples AI companies should follow. These agencies' focus on investigating incidents has made flying safer by learning from each accident.

“Accidents are going to happen. We're going to learn as much as we can and react from each one.”

Safety researchers warn of racing

Jacob Coxon, who quit Anthropic, accused AI labs of racing toward superintelligence and gambling with lives. Other researchers warn advanced systems could evade human control or enable catastrophic threats.

“Racing straight to self-improving superintelligence and gambling with our lives.”

Altman acknowledges doomsday risks

Altman said fears about AI are legitimate, citing risks of loss-of-control accidents and power concentration among a few developers. He argued the solution is not abandoning AI but ensuring safety stays ahead of capabilities.

“The world is right to be afraid of this.”

Companies must slow down if needed

Altman said AI companies should be prepared to slow down or stop if they cannot keep alignment and safety ahead of capabilities. The goal is preventing incidents from escalating through quick disclosure and learning.

Trump administration resists regulation

The Trump administration has shown little interest in regulating AI to maintain US competitiveness with China. David Sacks, a presidential advisor, told Altman and Anthropic's CEO to pace development without needing permission.