Cybersecurity Policy Report, U.K.’s Cyber Center Offers Guidance on Agentic AI Cyber Risks, (Aug 24, 2026)

By Tony Foley

The United Kingdom’s National Cyber Security Centre (NCSC) published interim practical guidance last week on managing the cyber risks associated with the proliferation of agentic AI (artificial intelligence), stressing the importance of considering how these systems behave when they don’t function as expected.

The guidance is aimed at system operators who are building environments where AI agents are intended to operate with significant degrees of autonomy or who are concerned about the implications of an agent performing unintended actions because of the instructions it receives, the tools available to it, or the systems it can access. It shares practical advice, based on NCSC research to date, to help organizations make informed decisions on how to deploy agentic AI systems securely within their environments.

NCSC suggests that organizations should assess how much autonomy their agentic AI systems need so that they may design proportional controls, understand a model’s built-in safeguards and protections, identify whether guardrails can be bypassed, and plan for additional safeguards.

Specifically, the guidance outlines the following key considerations in deciding whether an autonomous AI agent is appropriate given an organization’s risk tolerance and what safeguards may be needed to manage the associate risks:

1. Identify what could go wrong, by highlighting any “red lines” that agentic AI should stay within;

2. Prompt AI agents carefully, thinking about what the agent is designed to achieve, the actions it should be allowed to take, and when it should seek human approval prior to deployment;

3. Set the right level of oversight, depending on risk tolerance, from having a human “in the loop” approving actions before they are taken, to a human “on the loop” monitoring actions and intervening as necessary, to a human “out of the loop” approach where AI acts independently without any human review;

4. Control the agent’s environment with a robust sandbox that manages what resources can and cannot be communicated both locally and over a network;

5. Log, audit, and monitor agentic AI activity as part of security operations;

6. Make AI activity easy to attribute when communicating with third-party systems by ensuring that traffic originates from IP addresses that support reverse lookups and adding identifying headers to outbound communications to act as a form of watermarking; and

7. Establishing a process for emergency shutdowns when an incident is detected or reported.

News: InternationalLegislation DataSecurity AINews