Google DeepMind Launches $10 Million Global Research Initiative on Multi-Agent AI Safety

Google DeepMind has unveiled a major research funding program worth up to $10 million, developed in partnership with Schmidt Sciences, the Cooperative AI Foundation, ARIA, and Google.org, to advance safety research in multi-agent AI systems. The announcement addresses a rapidly evolving landscape where AI systems are becoming increasingly complex and interoperable. As AI technology scales, millions of AI agents created by different organizations are expected to increasingly interact within digital environments, exchanging information and conducting autonomous transactions. The core challenge lies in managing the unpredictable behaviors that emerge when independent AI agents interact at scale. Traditional safety evaluations examine models in isolation, but this approach fails to capture the complex, emergent dynamics that arise from multi-agent interactions. Researchers lack adequate tools to predict, measure, and monitor these collective behaviors, creating what experts term "invisible" safety risks. The funding call targets four priority research areas focused on understanding and mitigating risks from large-scale multi-agent coordination. Researchers worldwide can submit proposals before the August 8, 2026 deadline, with award announcements expected in autumn 2026.