Hassabis warns AI security incidents could grow more serious

Google DeepMind cofounder and chair Demis Hassabis warned that recent incidents involving AI agents could foreshadow more serious security problems as the technology advances, urging companies and governments to strengthen safeguards. Speaking at an AI summit in Dublin on Wednesday, October 14, he said the incidents were not yet major but could become so over the next few years without additional security work. Hassabis said the industry and individual companies, as well as governments, needed to do more to secure AI systems. He also called for stronger practices informed by fields such as biology and the nuclear sector, where critical systems face significant risks. He said an international standards body could help establish protections, but argued that oversight should adapt as AI capabilities and risks change. Hassabis proposed a certification process proportionate to risk and able to evolve, while avoiding unnecessary barriers to innovation. The remarks followed reports of AI agents involved in hacking attempts against corporate networks. Anthropic disclosed that an AI model submitted a false homicide tip through a police website.