Investing in multi-agent AI safety research

Google DeepMind, Schmidt Sciences, the Cooperative AI Foundation, ARIA, and Google.org are launching a 10 million dollar funding initiative to support global research into multi-agent AI safety. As AI technology evolves, millions of independent agents will soon interact, negotiate, and transact across digital environments. This program aims to address the risks that emerge when these autonomous systems operate in groups, rather than in isolation.

Current safety evaluations primarily focus on individual models, leaving a gap in understanding how large-scale agent populations behave. When many agents interact, they can develop complex, emergent behaviors that are difficult to predict or monitor. This research initiative seeks to identify and mitigate these systemic risks, such as unpredictable economic volatility or new security vulnerabilities, before they become widespread.

The funding call invites academic and independent researchers to submit proposals across four priority areas: creating realistic testing environments, studying the properties of agent networks, strengthening infrastructure for secure cross-platform interactions, and developing methods for oversight and control. By fostering a diverse, global research community, the initiative aims to establish transparent and robust safety standards for the future of AI. Interested researchers must submit their proposals by August 8, 2026, with award recipients expected to be announced in the autumn of 2026.