GOOGLE DEEPMIND SOUNDS ALARM: MILLIONS OF AI AGENTS ARE ABOUT TO INTERACT AND NOBODY KNOWS WHAT HAPPENS Google DeepMind has announced a $10 million research funding call targeting what it calls the coming multi-agent crisis. The concern is specific and it is serious. Right now AI agents built by different companies are beginning to interact with each other autonomously. They negotiate, communicate, transact, and execute tasks across systems without meaningful human oversight. The problem is that nobody designed this ecosystem and nobody fully understands what emerges when millions of agents start operating inside it simultaneously. Rohin Shah, who leads AGI safety and alignment research at DeepMind, described the risks plainly. Agents trained on different objectives, by different organizations, with different safety properties, are now sharing digital environments. The failure modes of individual agents are reasonably well studied. The failure modes of millions of agents in dynamic interaction with each other are almost entirely unknown. DeepMind is partnering with Schmidt Sciences, the Cooperative AI Foundation, and the Advanced Research and Invention Agency to fund researchers worldwide who can build the theoretical and empirical foundations needed to understand this problem before it becomes unmanageable. The application deadline is August 8. The funding is meaningful but the acknowledgment may be more important. One of the most powerful AI labs in the world is publicly saying it does not know what happens when its own technology starts talking to everyone else’s. Keywords: Google DeepMind multi-agent safety, AI agent risk research, agentic AI safety 2026, AI coordination risks