Five Eyes Issues Agentic AI Cybersecurity Guidance to Curb Autonomous-Action Risks

Agentic AI can do more than generate content: it can connect to tools, access data and execute tasks autonomously. If granted excessive privileges or compromised through prompt injection or identity impersonation, its erroneous actions could affect critical infrastructure. The Five Eyes alliance of the United States, United Kingdom, Canada, Australia and New Zealand therefore treats agentic AI as a distinct category of cybersecurity risk.

On May 1, 2026, the U.S. Cybersecurity and Infrastructure Security Agency and National Security Agency jointly issued guidance with the UK's National Cyber Security Centre, the Australian Cyber Security Centre, the Canadian Centre for Cyber Security and New Zealand's National Cyber Security Centre. The document grouped risks into 5 categories: permissions, design and configuration, behavior, structure, and accountability. It recommended starting with low-risk tasks, limiting access to context and permissions, and requiring human approval, comprehensive logging and continuous monitoring for high-impact or irreversible actions.