TORONTO, Canada – A senior OpenAI board member has warned that increasingly powerful artificial intelligence systems could escape human control within the next three years, potentially resulting in mass casualties if humanity loses the ability to manage the technology. Paul Christiano, who joined OpenAI’s Foundation Board and Safety and Security Committee in September, issued the warning while speaking at the Hinton Lectures in Toronto on Tuesday, November 10. He cautioned that so-called superhuman AI systems could eventually exceed humanity’s ability to understand, predict or control their behaviour.
Christiano said such systems could conceal their capabilities, hide actions that humans would not approve of and deliberately create confusion to avoid detection. He also warned that people could struggle to anticipate the consequences of following instructions generated by increasingly advanced AI. He pointed to a reported incident in which AI agents circumvented security restrictions during internal testing at OpenAI, gaining unauthorised access to parts of the company’s research infrastructure and systems operated by Hugging Face, an open-source AI development platform.
OpenAI subsequently acknowledged the incident, saying its models had exploited vulnerabilities in testing environments to bypass internet restrictions and access external systems. The company said it had since strengthened safeguards, improved monitoring and tightened security controls. Despite OpenAI introducing a framework to address cybersecurity threats and the risk of losing control over advanced AI systems, Christiano warned that existing safeguards remain inadequate to deal with the potential dangers.
His warning comes amid growing international efforts to improve AI oversight. On November 9, governments and leading technology companies, including OpenAI, Microsoft and Anthropic, signed the Paris Declaration on AI Safety and Security, committing to stronger cooperation, information sharing and early-warning systems to address emerging risks. However, concerns remain over the effectiveness of voluntary commitments without binding deadlines or enforcement measures.
Separately, Maxime Fournes, head of the PauseAI campaign group, warned that increasingly powerful AI could be used to launch devastating cyberattacks against banking networks, telecommunications systems and other critical infrastructure. Speaking to Anadolu Agency on November 9, Fournes estimated there was a 50% chance of a major AI-related cybersecurity catastrophe within the next year unless urgent action was taken.
