Crew Scaler Publishes Comprehensive Study on Multi-Agent AI Security

Crew Scaler announced in a press release the publication of a 120-page study titled "Security Considerations for Multi-Agent Systems," one of the first comprehensive security analyses focused on agentic AI. The nonprofit organization examined the risks and countermeasures for systems that deploy multiple autonomous AI agents capable of planning, delegating, and coordinating tasks.

The study evaluated 16 existing security and risk management frameworks against more than 1,000 distinct risk items across nine categories. Researchers found that current AI safety checklists are insufficient for multi-agent setups, identifying significant gaps in how organizations manage these systems.

Crew Scaler’s recommendations include granting the minimal tool authority necessary for each task, segmenting agent memory by workflow or tenant, treating inter-agent messages as untrusted input, monitoring for unpredictable behavior, and enforcing strict access controls to prevent data leaks. The research also suggests combining multiple security frameworks rather than relying on a single standard.

The full study is available at no cost on arXiv, and Crew Scaler invites organizations and researchers to collaborate on pilot projects applying the findings.