Scaling AI security research in a multi-agent world
Over the last decade, we have focused on improving the capabilities, usefulness and security of individual AI models. Today Google DeepMind – together with Schmidt’s teachings, Cooperative AI Foundation, Agency for Advanced Research and Inventionsand supported by Google.org — announces a up-to-date call for technical research funding of up to $10 million for researchers from around the world.
As AI technology scales, we are entering a up-to-date era. Soon, millions of AI agents – built by different organizations – will interact in digital environments, communicating, negotiating and transacting with each other.
When these systems interact, they must do so in a safe and sound and predictable manner. This change creates a significant opportunity: we can strengthen the security and stability of the entire AI ecosystem from the very beginning.
The funding call focuses on examining how large-scale, multi-agent AI systems behave as a group and how we can provide a framework for understanding and mitigating potential threats. By empowering researchers around the world, our goal is to solve the “invisible” security threats that arise when independent systems interact across different networks.
Why the agent ecosystem matters
When vast groups of AI agents interact, up-to-date collective behaviors and capabilities can suddenly emerge. We currently lack the tools to predict, measure and monitor these changes. Most security assessments analyze models in isolation. However, as We AND others As we argued earlier, interacting autonomic factors can produce complicated, “emergent” behaviors that are hard to predict.
Because this is a up-to-date area of research, it is critical to understand how these changes occur. For example, could they cause an unpredictable escalate in economic activity or lead to up-to-date security challenges? Our main goal is to understand how to manage these system-wide behaviors.
Scaling the frontiers of multi-agent security research
While basic multi-agent security frameworks exist, the rapid evolution of these systems requires immediate expansion of large-scale research.
Our Research 2025 we have established a framework for understanding these interactions, while our recent work on Pitfalls of AI agents examines the vulnerabilities agents face in hostile environments. Now we need to act faster. We are at a critical moment where the complexity of interactions between multiple agents exceeds existing security models.
The aim of this funding call is to accelerate progress by supporting a global network of independent researchers. A diverse community is vital to ensure safety standards are clear and resilient for all.
These efforts also advance the mission of Schmidt Sciences The science of trustworthy artificial intelligence AND AI agents programs supporting fundamental work in understanding and mitigating the risks associated with pioneering artificial intelligence systems, as well as ARIA Scaling trust program that aims to unlock up-to-date forms of cyberphysical multi-agent coordination.
A common call to action
No single lab can solve multi-agent security issues on its own. We invite academic and independent researchers to submit applications in four priority areas:
- Sandboxes and test environments: Create realistic, reproducible environments to assess, compare and accelerate progress in all areas of multi-agent security. This includes virtual marketplaces, simulated ecosystems, and cross-organizational workflows.
- The science of agent networks: Understanding the safety-relevant properties of populations of interacting agents, including examining how collective capabilities emerge and scale, how networks fail or become unstable, and how to detect risky, unexpected properties at the population level.
- Strengthening the agent infrastructure: Load testing of identity, reputation, and engagement protocols that constitute secure agent interactions across multiple platforms.
- Supervision and control: Developing methods to monitor populations of deployed agents and mitigate collective harm at scale.
How to participate
We invite researchers to review our call for proposals and join us in building a secure foundation for a multi-agent future.
The deadline for submitting applications is August 8, 2026, and the winners will be announced in fall 2026.
For more information on technical requirements and the application process, please visit our website application portal.