Google DeepMind Launches Up to $10M Multi-Agent AI Safety Research Funding Call
Summary
- • Google DeepMind and partners are offering up to $10M to researchers studying safety of large-scale multi-agent AI systems.
- • The funding targets emergent collective behaviors that arise when millions of AI agents from different organizations interact.
- • Four research priorities: sandboxes/testbeds, science of agent networks, agent infrastructure security, and oversight/control.
- • Partners include Schmidt Sciences, Cooperative AI Foundation, ARIA, and Google.org; proposals open to global researchers.
Details
Up to $10M in research funding offered for multi-agent AI safety
Google DeepMind and partners are offering up to $10 million in technical research funding for multi-agent AI safety researchers worldwide, via an open proposal call.
Coalition spans Google DeepMind, Schmidt Sciences, ARIA, and Google.org
Funding partners include Google DeepMind, Schmidt Sciences (Science of Trustworthy AI program), the Cooperative AI Foundation, ARIA (UK Advanced Research and Invention Agency, Scaling Trust programme), and Google.org.
Emergent behavior from agent interaction is an unsolved safety gap
When large groups of AI agents interact, new collective behaviors can emerge suddenly and unpredictably. Current safety evaluations only assess models in isolation, leaving systemic risks unmeasured and unmitigated.
Four priority research areas span evaluation, networks, infrastructure, and oversight
Proposals sought in: (1) sandboxes and testbeds for reproducible evaluation; (2) science of agent networks including failure modes; (3) strengthening agent infrastructure for secure cross-platform identity and reputation; (4) oversight and control methods for deployed agent populations at scale.
Initiative extends DeepMind's 2025 multi-agent safety framework and Agent Traps work
Google DeepMind's 2025 research established a framework for understanding multi-agent interactions; recent work on 'AI Agent Traps' explored vulnerabilities agents face in adversarial environments.
Open global call signals that no single lab can solve multi-agent safety alone
The call is open to global academic and independent researchers, reflecting a view that diverse, transparent participation is essential to build safety standards that work across different organizations and networks.
Financials = funding data; Partnership = coalition info; Research = scientific focus areas; Context = prior work; Insight = strategic framing
What This Means
As AI moves from individual models to large populations of interacting autonomous agents — potentially millions from different companies operating across shared digital environments — a new category of systemic risk emerges that existing safety tools aren't built to handle. Google DeepMind's $10M funding call, backed by a broad international coalition, represents one of the most concrete investments to date in addressing this frontier. The four priority research areas — from testbeds to population-level oversight — could lay the groundwork for safety standards that govern how AI agents communicate, transact, and coordinate as they proliferate. This initiative may shape regulatory thinking and industry norms around agentic AI deployment for years to come.
Sources
- Google DeepMind is worried about what happens when millions of agents start to interactMIT Technology Review
- Investing in multi-agent AI safety researchGoogle DeepMind