American And Allied Cyber Agencies Issue First Joint Guidance On Securing Agentic AI

What You Need to Know

Key takeaway #1

Agentic AI —artificial intelligence that can take actions across software or physical environments to achieve a specific objective with little to no human oversight — introduces security risks that exceed those of traditional software, including an expanded attack surface, the potential for compromised privilege, scope creep, behavior misalignment, and opacity in how agentic systems function.

Key takeaway #2

Organizations should adopt agentic AI through a deliberate application of best practices, including least-privilege access, rigorous monitoring, phased deployment, and mandatory human approval for high-risk actions.

Key takeaway #3

The regulatory environment for agentic AI is accelerating globally — including U.S. government procurement requirements, National Institute of Standards and Technology (NIST) AI Agent Standards Initiatives, the EU AI Act, and China’s new agentic AI policy — and organizations should expect increasing