Investing in multi-agent AI safety research

Google DeepMind and partners are announcing a $10M research funding initiative to address safety challenges in large-scale multi-agent AI systems where independent AI agents interact across digital environments.

• •Multi-agent AI systems can develop unpredictable emergent behaviors when millions of independently-built agents interact, negotiate and transact with each other
• •Current safety evaluation frameworks focus on individual models in isolation but lack tools to predict, measure and monitor collective behaviors at scale
• •Four research priority areas: building sandboxes and testbeds for multi-agent evaluation, understanding agent network safety properties, strengthening agent infrastructure protocols for secure interactions, and developing oversight methods for deployed agent populations
• •Research funding targets academic and independent researchers globally, with applications due August 8, 2026