Google DeepMind, Schmidt Sciences, the Cooperative AI Foundation (CAIF), and the UK’s Advanced Research and Invention Agency (ARIA) announced a $10 million technical research funding call on June 11, targeting the safety risks that emerge when large populations of AI agents interact across shared environments. Google.org is providing additional support.
The initiative addresses a gap that the partners describe as increasingly urgent: most AI safety evaluations today analyze models in isolation, but the industry is rapidly deploying systems where agents built by different organizations communicate, negotiate, and transact with one another. According to DeepMind’s announcement, “interacting autonomous agents can produce complex, ‘emergent’ behaviors that are difficult to anticipate,” and the complexity of these multi-agent interactions is “outpacing existing safety models.”
Funding Structure
The Cooperative AI Foundation detailed two funding tiers: Tier 1 grants of up to $300,000 and Tier 2 grants of up to $1,000,000, available to independent and academic researchers worldwide. Applications close August 9, 2026 (Anywhere on Earth), with awardees expected in Autumn 2026.
Four Research Priorities
The call targets four specific areas:
- Sandboxes and testbeds. Building reproducible environments to evaluate multi-agent safety, including virtual marketplaces, simulated ecosystems, and multi-organization workflows.
- The science of agent networks. Understanding how collective capabilities emerge and scale, how networks fail or become volatile, and how to detect dangerous population-level properties.
- Strengthening agent infrastructure. Stress-testing protocols for identity, reputation, and commitment that secure cross-platform agent interactions.
- Oversight and control. Developing methods to monitor deployed agent populations and mitigate collective harms at scale.
Why the Timing Matters
The funding call builds on DeepMind’s prior work in this space. The lab’s 2025 research established foundational frameworks for understanding multi-agent interactions, and its AI Agent Traps paper explored vulnerabilities agents face in adversarial environments. CAIF noted that while “some of these problems will be addressed by market forces, we expect others to fall through the gaps,” according to the foundation’s announcement.
The partnership structure itself signals scale of concern. Schmidt Sciences’ involvement connects to its Science of Trustworthy AI and AI Agents programs. ARIA’s participation links to the UK agency’s Scaling Trust programme, which focuses on “new forms of cyber-physical multi-agent coordination,” per DeepMind. Google.org rounds out the consortium as its charitable arm.
The Infrastructure Gap
The four research areas map directly to unsolved problems in production agent deployments. Agent identity and reputation protocols, for example, remain fragmented. OpenClaw, Microsoft’s Agent 365, and Google’s own Antigravity platform each handle agent identity differently, with no cross-platform standard for verifying which agent is acting on whose behalf. The “strengthening agent infrastructure” priority targets exactly this gap.
The oversight and control area addresses what happens when thousands of agents from different principals operate on the same infrastructure simultaneously. Monitoring individual agent behavior is a solved (if imperfect) problem. Monitoring emergent behavior across a population of agents, each with different goals and principals, is not.
Researchers interested in applying can find the full technical requirements and application portal through the Cooperative AI Foundation’s grants page.