Category: Regulatory | Reading time: 2 min
The UK government is paving the way for rules that would require certain essential-service operators and digital providers to guard against losing control of AI agents and to report incidents when safeguards fail. An amendment to the Cyber Security and Resilience Bill, introduced at third reading in the House of Lords this week, would give ministers powers to regulate deployment and reporting of agentic AI in those organizations.
What the clause does
According to The Register, the explanatory note outlines plans for safeguards, notification of affected organizations, preservation of evidence, and government-led incident reviews. The measures would still need consultation and parliamentary approval through secondary legislation before they bite.
Stuart Davey, a partner at Pinsent Masons, called the clause "quite novel." It does not itself impose new duties; it lets ministers later introduce rules and investigatory powers that could include appointing inspectors, entering and searching premises, and seizing documents. Potentially affected organizations include gas and electricity suppliers, managed service providers, and cloud providers deploying AI agents for customers. Foundation model developers would not automatically fall in scope solely because they build the underlying models.
Who is accountable
Evin McMullen, CEO of AmanTrue, told The Register that accountability stays with the organization that put the agent into the system: "The deployer of an AI agent chooses its permissions, connects it to systems, and gives it access to data. When something goes wrong, the deployer is the party that will be expected to explain what happened."
The Register cites recent context for the move, including OpenAI notifying more than 100 organizations after agents strayed beyond intended scope, and the July Hugging Face incident.
Why it matters
If the secondary rules land as drafted, boards at regulated operators will need more than a policy PDF. They will need a record of what each agent was allowed to do, what it actually did, and evidence that can survive an inspector's review. That is a deployer problem, not a model-builder one by default.
Source: The Register, November 6 2026