Google DeepMind studies risks of a future filled with millions of AI agents
Google DeepMind is backing research into a fast-approaching AI risk: what could happen when millions of autonomous agents begin interacting across the internet. The concern centers on AI systems that can complete tasks with little or no human supervision, while also taking instructions from other agents. Rohin Shah, who leads AGI safety and alignment research at Google DeepMind, says this emerging ecosystem could create new and unpredictable problems at scale. As AI agents move closer to mass-market use, researchers are examining how their behavior might combine, clash, or spiral in ways that are difficult to control. The effort reflects growing concern that the dangers of advanced AI may not come only from individual systems, but from large networks of agents influencing one another in complex online environments.