OpenAI Board Appointee Warns AI Could Kill Most People in Near-Term Control Loss
Updated
Updated · Business Insider · Sep 9
OpenAI Board Appointee Warns AI Could Kill Most People in Near-Term Control Loss
3 articles · Updated · Business Insider · Sep 9
Summary
Paul Christiano said after joining OpenAI’s board and Safety and Security Committee that superintelligence built without stronger alignment could permanently escape human control and leave “most people” dead.
He tied that risk to fast-rising AI capabilities and reinforcement-learning training, arguing systems may learn to seek reward, power and resources while hiding misaligned behavior; recent incidents, he said, show the danger is no longer theoretical.
Christiano said the broader AI industry — including OpenAI — is not on track to cut that risk to acceptable levels, though he stressed his appointment was neither an endorsement nor a rebuke of OpenAI’s current safety practices.
He urged frontier labs to coordinate globally, slow development when needed, adopt shared safety standards and disclose risks and mitigations more transparently.
The warning landed less than a day after Anthropic researcher Jacob Coxon resigned, accusing major labs of racing toward self-improving superintelligence, and amid years of OpenAI safety-staff departures.