A new chapter in artificial intelligence is unfolding, and with it, a critical call to action for global researchers. Today, a powerful coalition comprising Google DeepMind, Schmidt Sciences, the Cooperative AI Foundation, the Advanced Research and Invention Agency (ARIA), and supported by Google.org, is committing up to $10 million towards groundbreaking research. This significant funding initiative aims to propel multi-agent AI safety research forward, addressing the complex challenges posed by increasingly interconnected AI systems.
For years, the focus of AI development has largely been on refining individual models, making them smarter, more capable, and safer in isolation. However, as AI technology rapidly scales, we’re on the cusp of an era where millions of independent AI agents, developed by various organizations, will interact seamlessly across digital landscapes. These agents will communicate, negotiate, and transact with one another, creating an intricate web of digital interactions.
Safeguarding the Multi-Agent AI Ecosystem
This paradigm shift presents both immense opportunities and novel safety concerns. When large groups of AI agents interact, new collective behaviors and capabilities can emerge unexpectedly, often without clear precedent. Crucially, we currently lack comprehensive tools to predict, measure, or effectively monitor these emergent transitions.
Most existing AI safety evaluations analyze models in isolation, a method that falls short in understanding complex system-wide dynamics. As a result, the “invisible” safety risks that arise from these intricate interdependencies remain largely unexplored and unmitigated. This research initiative seeks to fill that void, building foundational frameworks to understand and manage these critical interactions.
Consider the potential ramifications: could emergent AI behaviors trigger unpredictable economic fluctuations, or even introduce entirely new security vulnerabilities? The core objective of this funding call is to equip the global research community with the insights and tools needed to anticipate and manage these system-wide behaviors, ensuring a stable and predictable AI future.
Accelerating Global Collaboration and Innovation
While foundational frameworks for multi-agent safety are beginning to emerge, the rapid evolution of AI systems demands an immediate and large-scale expansion of dedicated research. Previous work by Google DeepMind, including a 2025 framework for understanding AI interactions and recent insights into “AI Agent Traps” in adversarial environments, has underscored this urgency. We are at a critical juncture where the complexity of multi-agent interactions is quickly outpacing our current safety models.
No single laboratory or organization possesses all the answers when it comes to multi-agent safety. This funding call is designed to cultivate a robust, global network of independent researchers, fostering a diverse community essential for developing transparent and universally applicable safety standards. This collective effort is vital for the benefit of everyone involved in the AI ecosystem.
This initiative also strongly aligns with the missions of the supporting partners. Schmidt Sciences’ Science of Trustworthy AI and AI Agents programs are dedicated to mitigating risks from frontier AI, while ARIA’s Scaling Trust program aims to unlock new forms of multi-agent coordination. Together, these organizations are committed to building a secure and trustworthy AI future.
Join the Effort: Submit Your Research Proposal
We extend an open invitation to academic institutions and independent researchers worldwide to contribute to this pivotal field. Proposals are welcomed across four priority areas, detailed further in our official call for proposals. This is a unique opportunity to shape the future of AI safety and contribute to a secure foundation for a multi-agent world.
The deadline to submit applications is August 8, 2026, with awardees expected to be announced in the Autumn of 2026. For comprehensive details on the technical requirements and the application process, please visit our dedicated application portal. Your expertise can help us build a safer, more predictable, and more responsible multi-agent AI future.
Source: Google DeepMind Blog
