The U.S. Center for AI Standards and Innovation (CAISI/NIST) issued a Request for Information on securing AI agent systems, and the Computing Community Consortium (CCC) and Computing Research Association (CRA) submitted a joint response. Key challenges identified include delegating credentials to AI agents, determining accountability for security failures, and preventing cascade effects in multi-agent systems. Recommendations include reevaluating fuzz testing and input sanitization for autonomous agents, researching effective guardrails, assessing multi-agent risks, and shifting toward security-by-design rather than post-deployment patching. The response also emphasizes the importance of human-in-the-loop practices and stronger collaboration between research communities and industry cybersecurity practitioners.
Sort: