A robotic hand reaching towards a glowing network pattern on a blue background

Your AI agents, your responsibility: what UK regulators want CAIOs to hear before 20 November

In the space of four weeks, three separate parts of the UK system have said much the same thing about AI agents. The Information Commissioner's Office has opened a call for evidence on agentic AI and data protection. The four regulators in the Digital Regulation Cooperation Forum have put out a joint statement on agents. And the House of Lords has added a clause on "loss of control" of AI agents to the Cyber Security and Resilience Bill.

None of these is a new rulebook yet. Together, though, they give Chief AI Officers a clear signal about where scrutiny is heading, and a practical deadline: the ICO's call for evidence closes at the end of Thursday 20 November 2026.

What has happened

The ICO wants to hear from organisations using agents. On 8 October the ICO launched a six-week call for evidence asking developers, deployers and other experts how they are managing the data protection risks of agentic AI. The questions cover data security, transparency, accountability, automated decision-making, fairness and purpose limitation, and the lawful basis for processing. The ICO says the answers will shape its future guidance and its planned statutory code of practice on AI and automated decision-making. Richard Nevinson, the ICO's Director of Technology Regulation, summed up the tone: "the fact AI agents act with autonomy is not an excuse for poor compliance."

The DRCF regulators say existing law already applies. On 30 October the Competition and Markets Authority, the Financial Conduct Authority, the ICO and Ofcom published a joint statement on personal AI agents. It is mainly about agents that act for consumers, such as making purchases or dealing with businesses. Its central line applies to any organisation deploying agents: "An AI agent's autonomy does not diminish the responsibility of organisations that deploy it to comply with applicable law and relevant regulation." The regulators also list the questions they expect firms to think about, including how people know when an agent is acting for them, how authority to act is given, how personal data is used, who is accountable for the agent's decisions and how people get redress when something goes wrong.

Parliament is preparing for agents that go wrong. The Cyber Security and Resilience (Network and Information Systems) Bill, which updates the UK's cyber rules for essential and digital services, came back from the Lords with a new clause headed "Functions of the Secretary of State: loss of control of artificial intelligence agent". It would let future regulations give the government powers where a security or operational compromise "is or may be attributable to loss of oversight or control of an AI agent", or where there is a risk of one. Those powers could include investigating, preserving evidence, testing fixes and requiring regulated organisations to act. What counts as an "AI agent" and as "loss of oversight or control" is left to later regulations. The Bill is now back in the Commons for MPs to consider the Lords' changes, so this is not law yet.

Why it matters for CAIOs

The common thread is accountability. Regulators are not asking whether an agent was clever enough to make its own choices. They are asking whether the organisation that deployed it knew what it could do, set limits, watched what happened and could put things right.

That lands squarely on the CAIO's desk. In many organisations, agent pilots have grown faster than the paperwork around them. A tool that started as a test in one team can end up with access to customer records, email and payment systems. If a regulator or an auditor asked you tomorrow to describe every agent in production and what it is allowed to do, how long would the answer take?

Six things to do now

Keep a register of your agents. For each one, record its owner, its purpose, the data it can read, the systems and tools it can call, and whether it can act without a person approving each step. Treat this like any other asset register and keep it current.
Write down each agent's authority to act. Spending limits, the kinds of message it can send, what it can change and what always needs a human. The DRCF statement names "how authority to act is given" as a key question. Have an answer on paper.
Make oversight something you can prove. The Bill's language is about "loss of oversight or control". Make sure agent actions are logged, that unusual behaviour raises an alert, and that someone can switch an agent off quickly. Test that switch.
Revisit your DPIAs and automated decisions. Where an agent makes or shapes decisions about people, check your data protection impact assessments, your lawful basis and how a person can ask for human review. These are the areas the ICO is asking about.
Tell people when an agent is acting. Customers and staff should be able to tell when they are dealing with an agent, and know how to reach a person. Plain notices now are cheaper than redress later.
Decide whether to respond to the ICO. If your organisation runs agents at any scale, your experience is exactly what the call for evidence is after, and responding is a chance to shape the guidance you will later have to follow. The ICO notes that it may publish organisations' responses in full or in summary, so agree with legal and communications what you are happy to share. Responses go through the ICO's Citizen Space portal and must be in by the end of 20 November.

The bottom line for CAIOs

The direction of travel is clear even though the detailed rules are still to come. Data protection, consumer, competition and cyber regulators all expect the organisation behind an agent to answer for it. The CAIOs who will find the next year easiest are the ones who can already show a register, clear limits, working oversight and a way for people to complain. If you have those, responding to the ICO by 20 November is a straightforward way to make sure your view is heard.