Google DeepMind, together with Schmidt Sciences, the Cooperative AI Foundation, and the Advanced Research and Invention Agency, has announced a research funding call of up to $10 million dedicated to multi-agent AI safety. Supported by Google.org, the global initiative will provide grants to independent researchers studying how large-scale populations of AI agents behave when interacting across digital networks.
Google DeepMind stated that while safety research over the past decade focused on individual AI models, technology is shifting toward environments where millions of autonomous agents built by different organizations will communicate, negotiate, and transact with one another. The organization noted that existing safety evaluations analyze models in isolation, leaving gaps in predicting emergent group behaviors, unexpected flurries of economic activity, or new security risks.
Priority research areas
The funding call invites academic and independent researchers to submit proposals across four technical focus areas:
- Sandboxes and testbeds: Building realistic environments such as virtual marketplaces, simulated ecosystems, and multi-organization workflows for reproducible safety evaluations.
- The science of agent networks: Investigating population-level properties, how collective capabilities emerge, and how agent networks become volatile or fail.
- Strengthening agent infrastructure: Stress-testing cross-platform protocols that manage identity, reputation, and commitment during agent interactions.
- Oversight and control: Developing methods to monitor deployed agent populations and mitigate collective harms at scale.
Program background and timeline
The program builds on Google DeepMind's 2025 research on agent interaction frameworks and its recent work on AI Agent Traps, which evaluated agent vulnerabilities in adversarial environments. The initiative also aligns with Schmidt Sciences' Science of Trustworthy AI and AI Agents programs, as well as the Advanced Research and Invention Agency's Scaling Trust programme, which focuses on cyber-physical multi-agent coordination.
Researchers worldwide have until August 8, 2026, to submit proposals through the program portal. Google DeepMind expects to announce the awarded research proposals in Autumn 2026.
