Four UK Regulators Warn Companies That Blaming Their AI Agent Will Not Work
The CMA, FCA, ICO and Ofcom say existing law already holds businesses responsible for what their AI agents do, though the joint statement creates no new rules. The sharpest legal line is older than the headlines suggest.
If an AI agent you hired to switch your broadband or cancel your subscriptions botches the job, Britain's regulators want you to know who answers for it. Not the model. Not the vendor. The business that put the agent in front of you.
On Friday the Digital Regulation Cooperation Forum, the body that links the Competition and Markets Authority, the Financial Conduct Authority, the Information Commissioner's Office and Ofcom, published a joint statement on personal AI agents. These are the assistants that can make purchases and deal with companies on your behalf. The regulators did not propose a rule. They did something blunter and cheaper: they said the rules already exist.
"An AI agent's autonomy does not diminish the responsibility of organisations that deploy it to comply with applicable law and relevant regulation." DRCF joint statement
That line, as quoted by law firm Burges Salmon, is the whole story in one sentence. The statement lists competition, consumer and data protection law, online safety, financial services, payments and cyber security requirements as already applying to agents. Where the regulators "reasonably suspect" a breach that risks harm to consumers or businesses, they say they "will not hesitate to use" their powers.
Old guidance, new megaphone
Here is the part the headlines may skip. The sharpest legal claim was not born on Friday. The CMA's business guidance, Using AI agents: complying with consumer law, was first published on 9 March 2026. Its Friday update, according to the page's change log, was a note pointing to the joint statement.
The guidance says it plainly: "you are responsible for what an AI agent does in the same way you are responsible for what an employee does." That holds even if someone else designed or supplies the agent. Break consumer protection law and the CMA says a business could be fined up to 10% of its worldwide turnover and possibly forced to compensate affected consumers. That figure is the general ceiling the CMA cites, not a penalty aimed at any company.
So what is actually new? Mostly the choir. Four regulators now sing the same line, and a single agent can touch all of their patches at once. An agent that moves money, uses personal data and nudges you toward a product could in theory draw attention from several of them.
Market and Tech reports the statement creates no new law, technical standard or certification scheme, and sets no specific tests or timetable. That matters. Nobody should read this as a UK ban or a licensing regime for agents. It is guidance and a promise of coordination, intelligence sharing and regulatory sandboxes.
The CMA is also keen to sound unthreatening. In its companion paper, Agentic AI and consumers, it says "we don't believe that heavy-handed, sweeping regulation is right for the UK" because it could stymie innovation. It also says most deployments today are bounded, such as customer service.
The paper still reads like a checklist of things that could go wrong. It flags agents that are not a "faithful servant" and steer people toward profitable products, hallucinations with costly real-world consequences, and what it calls agentic collusion, where competing businesses' pricing agents dampen competition. Its worked example, an agent that switches providers and cancels subscriptions, recommends mandatory confirmation for high-risk actions, strong audit logs and clear liability if the agent acts outside customer instructions.
The pressure is building from other directions too. The ICO said on 8 October it had made enquiries with OpenAI, Anthropic, Meta and the UK's AI Security Institute about recent agentic AI testing, after reports that some agents bypassed protections. Meanwhile platforms are choosing sides on access, as we reported when big banks started locking out AI agents.
My read: the useful signal here is for anyone buying an agent from a third party. The contract may move the blame on paper. According to the CMA, it does not move it in law. Expect procurement teams to start asking vendors awkward questions about logs, testing and who pays when the agent gets it wrong.
