CISA and Five Eyes partners publish Careful Adoption of Agentic AI Services guidance

CISA, the NSA and international partner cybersecurity agencies published joint guidance for securing agentic AI systems on May 01, 2026. The document, Careful Adoption of Agentic AI Services, lays out cybersecurity risks tied to autonomous AI agents and recommends mitigations for organizations that build, deploy or operate them.

What it is

The guide was co-authored by CISA, the NSA, the Australian Signals Directorate's Australian Cyber Security Centre, the Canadian Centre for Cyber Security, New Zealand's NCSC-NZ and the UK's NCSC-UK.

NSA described the publication as a comprehensive guide to understanding and mitigating the unique risks associated with agentic AI within critical infrastructure, including the defense sector.

The agencies noted that critical infrastructure and defense sectors are increasingly deploying agentic AI systems to support mission-critical systems and capitalize on automation benefits.

Unlike traditional generative AI, which typically requires human validation, agentic AI systems are designed to operate autonomously, which the agencies said makes them a powerful tool with distinct risks.

What it requires

The guidance warns that agentic systems can introduce additional cybersecurity risks, such as an expanded attack surface, privilege creep, behavioral misalignment, and obscure event records.

NSA's summary breaks the risks into five categories, including that over-privileged agents can amplify the impact of a single compromise.

Other categories include design and configuration risks where insecure design and provisioning can introduce vulnerabilities, and structural risks from the interconnected nature of agentic systems.

The authoring agencies strongly recommend aligning agentic AI risks and mitigation strategies with an organization's existing security model and risk posture.

What security leaders should do

CISA's recommended actions include avoiding granting broad or unrestricted access, especially to sensitive data or critical systems.

Organizations are also encouraged to begin with agentic AI use cases that are low-risk and non-sensitive before expanding scope.

The ASD-led text adds that organisations should only use agentic AI for low-risk and non-sensitive tasks as they build confidence in the technology.

NSA's summary recommends deploying agentic AI incrementally, continuously assessing against evolving threat models, and maintaining strong governance, explicit accountability, rigorous monitoring, and human oversight as essential rather than optional.

Security leaders evaluating agentic deployments should map these five risk categories against their own agent inventories, confirm least-privilege access, and verify that logging captures agent decisions before autonomy is expanded further.