Google DeepMind studies risks from millions of interacting AI agents

Google DeepMind is funding research into the safety risks that may appear when millions of AI agents begin interacting with each other online.

That is a different problem from testing one chatbot in isolation. Agents can negotiate, delegate, manipulate markets, overload systems, or amplify each other’s mistakes in ways that are hard to predict from single-agent benchmarks.

The research reflects a growing concern that agent safety will need tools for ecosystem-level behavior, not just model-level alignment.