Demis Hassabis Warns Weak AI Safeguards Could Lead to Major Incident

Demis Hassabis, chairman of DeepMind and Google’s chief scientist, warned that inadequate safeguards could allow AI systems to trigger a major security incident next year or in the years that follow.

At an AI summit, Hassabis described the past year as a shift toward AI agents, systems that can take actions on their own. He said that autonomy can make AI more useful for real-world tasks, while also increasing its potential risks.

Hassabis pointed to AI attacks this year as an early warning, while saying current systems have not yet been powerful enough for incidents to cause major consequences. He urged the industry to build stronger safeguards, drawing on security practices in fields such as biology and nuclear technology. He said substantial work remains to put protections in place.

Hassabis also called for international cooperation and proposed creating an international standards body to set safety requirements for AI systems.