Google DeepMind opens $10M research call on multi-agent AI safety

Google DeepMind on June 11, 2026 opened a research funding call of up to $10 million focused on the safety of large-scale multi-agent AI systems, the lab said in a blog post. It is running the program jointly with Schmidt Sciences, the Cooperative AI Foundation and the United Kingdom's Advanced Research and Invention Agency (ARIA), with support from Google.org.

The call targets a gap that grows as companies deploy more autonomous agents: the failure modes that appear when agents transact, negotiate and coordinate with one another rather than acting alone. DeepMind is asking researchers to work across four areas — sandboxes and testbeds, the science of agent networks, agent infrastructure such as identity, reputation and cross-platform security, and the oversight and control of deployed agent populations.

Applications close on August 8, 2026, and awardees will be announced in autumn 2026, according to the program details; submissions run through the Schmidt Sciences application portal. The lab pointed to its own 2025 paper on a multi-agent framework and a separate paper on "AI Agent Traps" as the technical groundwork for the effort.

The initiative folds into existing bets by the partners. It aligns with Schmidt Sciences' Science of Trustworthy AI and AI Agents programs and ARIA's Scaling Trust programme, the organizations said.

The money is modest against frontier-model training budgets, and a funding call does not by itself resolve how agent networks behave at scale — it pays for the research that might. How many agents interact, under what protocols and with what oversight remains largely untested in production, and the four priority areas read as an admission that the tooling to study those interactions does not yet exist.

For researchers, the August deadline is the near-term signal to watch. For the wider industry, the call is a coordinated attempt to build safety scaffolding before multi-agent deployments become routine.