AI agents that defy orders: Britain proposes inspection, testing and seizure powers

What should happen when an AI agent stops obeying the company that uses it? The United Kingdom gave its answer this week. During the third reading of the Cyber Security and Resilience Bill in the House of Lords, ministers spoke at length about rules to handle the risks posed by agentic AI - systems that act directly on files, software and services rather than merely answering questions.

Under the proposal, a framework is being created to give the relevant authorities powers of inspection, testing and seizure over AI agents that act beyond their instructions or slip out of control. Every organisation that allows software agents to operate inside an electricity supplier, a cloud platform or a managed service provider needs to watch this change.

Why it matters: as agentic AI moves quickly into real enterprise environments, this is one of the first attempts to set out a legal approach for the moment an agent loses control. Organisations that deploy AI agents should review their governance arrangements now.