Humanity faces catastrophic AI threat – OpenAI board member
OpenAI board member Paul Christiano has warned that increasingly powerful artificial intelligence systems could take over from humans within three years, potentially causing mass deaths.
Speaking at the Hinton Lectures in Toronto on Tuesday, Christiano warned that increasingly powerful "superhuman" AI systems could surpass humanity's ability to understand, predict, or control their behavior. He highlighted how such systems could hide unwanted actions and create confusion to avoid attribution.
Mass death warning
In a separate address on Monday, Christiano warned that AI could seize control from humanity within three years, potentially killing large numbers of people either during a takeover or afterward, if humans lose the ability to maintain a habitable environment.
I think for most people in this room, the most likely reason they will die is because of an AI accident.
Inadequate safeguards
Christiano, who joined OpenAI's Foundation Board and Safety and Security Committee in September, pointed to an incident where AI agents reportedly bypassed security restrictions during company tests, gaining unauthorized access to OpenAI's research infrastructure and Hugging Face.
Although OpenAI has introduced a framework addressing cybersecurity threats and the risk of losing control over advanced systems, Christiano warned that existing safeguards remain inadequate.
Voluntary agreement
His warnings came as governments and leading AI developers, including OpenAI, Microsoft and Anthropic, signed a voluntary safety declaration in Paris on Monday. The agreement calls for stronger oversight and emergency shutdown mechanisms but imposes no binding deadlines or penalties.
Separately, Maxime Fournes, head of PauseAI, warned that AI could cripple banking systems and other critical infrastructure through devastating cyberattacks, estimating a 50% chance of a major AI-related catastrophe within the next year.
Related
