Crew Scaler Releases Landmark Agentic AI Security Study for Multi-Agent Systems

BALTIMORE, May 26, 2026 /PRNewswire/ -- Crew Scaler, an AI-first nonprofit focused on safe and secure AI adoption, has released one of the first book-length security analyses of multi-agent "agentic AI" systems. The new paper, "Security Considerations for Multi-Agent Systems," outlines concrete risks and countermeasures for organizations deploying multiple autonomous AI agents in production environments.

Unlike traditional chatbots, agentic AI systems do not just answer questions — they plan, delegate, use tools, retain memory, and coordinate across workflows. "Agentic AI is where many organizations expect their real productivity gains to come from — but those same systems introduce whole new failure modes," said Tam Nguyen, CEO of Crew Scaler and a Senior AI and security expert in the U.S. government. "Our goal with this research is to give security teams, architects, and policymakers a practical map of the risks, not just abstract principles, so they can move forward with confidence instead of guesswork."

In the study, Crew Scaler researchers evaluated 16 security and risk management frameworks against more than 1,000 distinct multi-agent risk items across nine categories. The findings are clear: significant gaps remain. The conclusion: traditional AI safety checklists are necessary but not sufficient for multi-agent systems.

The paper translates its analysis into practical recommendations for any organization deploying multi-agent systems, including but not limited to: minimal tool authority needed for each task; segmenting memory by workflow, team, or tenant; treating inter-agent messages as untrusted input; monitoring for non-deterministic behavior and unexpected tool chains; preventing data leakage with strict access controls; and combining multiple security frameworks rather than relying on one standard.

At more than 120 pages, the study provides one of the most comprehensive publicly available treatments of multi-agent security and contributes to ongoing policy and standards efforts in AI risk management. The full paper is available at no cost at https://arxiv.org/abs/2603.09002.

Organizations and researchers interested in applying the findings or collaborating on pilot projects can learn more at https://crewscaler.org.

About Crew Scaler

Crew Scaler is a nonprofit organization dedicated to helping communities, workers, and small organizations adopt AI safely and securely. The organization combines research, training, and hands-on advisory work to close the gap between cutting-edge AI systems and real-world governance and security practice.

About the Authors

Tam Nguyen is CEO of Crew Scaler and a Senior AI and security expert in the U.S. government. Moses Ndebugre is a Senior Researcher at Crew Scaler and a PhD candidate in Electrical and Computer Engineering at NC A&T. Dheeraj Arremsetty is an AI Technical Solution Architect at IBM and an Advisory Board Member at Crew Scaler.

Media Contact:
Tam Nguyen (CEO Crew Scaler)
+1 (970) 404-1232
[email protected]
https://crewscaler.org

SOURCE Crew Scaler