Google DeepMind funds research on the risks of millions of interacting AI agents

Google DeepMind is putting about $10 million toward research into the risks of large-scale multi-agent systems — what happens when millions of AI agents transact, negotiate, and act online with little or no human in the loop, according to MIT Technology Review.

The framing is a tell about where the industry thinks deployment is headed. Most AI safety work has focused on a single model’s behavior; DeepMind is funding the next problem — emergent dynamics like collusion, cascading failures, and feedback loops that only appear once many autonomous agents interact at scale. It is an early acknowledgment that agent ecosystems may need governance distinct from the models that power them.

Read the source →