Investing in multi-agent AI safety research

Google DeepMind and partner organizations are announcing a funding call for research into the safety of multi-agent AI systems, which refers to multiple AI agents built by different organizations interacting with one another across digital environments. As AI technology scales, these systems will communicate, negotiate, and transact with each other, creating new collective behaviors that are difficult to predict using current safety evaluation methods that typically analyze individual models in isolation. The funding aims to support researchers globally in understanding how to manage system-wide risks and emergent behaviors that arise when independent AI agents interact, focusing on areas like building test environments for multi-agent evaluation and studying how agent networks function.