Scaling AI Safety for a Multi-Agent World
This grant funds research advancing the safety, oversight, and control of large-scale, multi-actor AI agent ecosystems through rigorous system-level scientific inquiry and technical innovation.
Funder: Schmidt Sciences
Due Dates: August 8, 2026 (Full application, 11:59 PM AoE) | August 9, 2026 (Application cutoff, 07:59 AM EDT)
Funding Amounts: Tier 1: Up to $300,000 (1–2 years) | Tier 2: $300,000–$1,000,000 (1–2 years)
Summary: Funds research to advance safety, oversight, and control of large-scale, multi-actor AI agent ecosystems through system-level scientific inquiry and technical innovation.
Key Information: For-profit entities cannot lead projects; global nonprofit and academic institutions are eligible.
Description
This opportunity supports foundational research to understand, evaluate, and control risks arising from large-scale ecosystems of interacting AI agents deployed by multiple actors. As AI agents proliferate in multi-agent and multi-principal settings, new safety challenges and collective risks emerge that cannot be addressed by focusing solely on individual models. The program seeks deep, targeted research that advances system-level understanding of multi-agent AI safety, emphasizing empirical rigor, scientific quality, and real-world applicability.
The research agenda is organized into four clusters:
- Sandboxes and Testbeds: Develop reproducible, realistic environments for empirical study of multi-agent systems.
- Science of Agent Networks: Analyze collective behaviors, vulnerabilities, and emergent properties in agent populations.
- Strengthening Agent Infrastructure: Advance technical infrastructure and protocols for trustworthy multi-agent interactions (e.g., identity, reputation, commitment mechanisms).
- Multi-Agent Oversight and Control: Create scalable methods for oversight, detection, attribution, and control of agent populations.
Proposals may target one or multiple clusters, but depth is prioritized over breadth. Out-of-scope topics include single-agent safety, capability advancement without a clear safety motivation, policy advocacy, and commercial product development.
