Rising risks from agentic AI

Hassabis said AI agents present new safety challenges beyond those of conventional language models. Among the scenarios he flagged are agents involved in hacking activities and agents submitting false or fabricated reports, which could mislead institutions and individuals. He argued that identifying such warning signs early is critical to preventing larger-scale incidents as agent deployments expand.

Call for early warning of major incidents

The DeepMind chief emphasized the need for systems that can detect and flag early signals before a major AI incident occurs, rather than responding after harm has spread. The warning reflects growing attention across the AI industry to the safety implications of autonomous agents that can act, communicate and make decisions with limited human oversight.

Uncertainty around timeline

The remarks underscore concern within DeepMind about how quickly agentic capabilities are advancing, though specific timelines or thresholds for what would constitute a "major incident" were not detailed in the available material.

As AI agents gain the ability to act autonomously online, Hassabis's warning about hacking and false reporting stands as an early call for vigilance, and the claim is judged True.

Verdict: True