Four UK regulators have written to ministers warning that personal AI agents are arriving in the UK market faster than the firms behind them are willing to explain what those products do. The joint letter, dated 16 November, comes from the CMA, the FCA, the ICO and Ofcom, and it names three risks the regulators want addressed before launch rather than after.
The letter says engagement from developers "has not been consistently early or open across the relevant regulators", while acknowledging that some firms have cooperated properly.
What the regulators are worried about
The four bodies, working through the Digital Regulation Cooperation Forum, single out three specific harms:
- Fraud and scams, where an agent is manipulated or impersonated to move money or data
- Personal information shared outside users' expectations, which in practice means an agent handing your data to a partner you never agreed to
- Misleading recommendations, decisions or information, where an agent acts confidently on something it has got wrong
The letter also questions whether the public understands how these products work at all, and asks that users keep control of their interactions with clear explanations of the risks specific to agentic systems.
The part that catches businesses out
The expectations are not limited to the companies building the agents. Banks, insurers and telecoms firms considering commercial partnerships are told their own obligations continue to apply when customers use a personal AI agent. That puts a bank integrating a third-party agent in the same conversation as the developer.
The letter also says the regulators will use their powers where they reasonably suspect breaches of the law that risk harm to consumers or businesses.
Why this matters if you use one
A personal AI agent is the category of product that books, buys and cancels on your behalf. The appeal is that you stop filling in forms. The risk is that it acts with your authority but not necessarily your judgement, and the failure is invisible until a payment has gone through or a subscription has renewed.
The regulators are not proposing a new law here. They are setting out how existing consumer, data protection and financial rules apply to a product category that has been launched largely without that conversation. That matters because the answer to "who is responsible when the agent gets it wrong" is now being established in advance rather than in a test case.
What to do before you connect one
- Read what the agent can access, not what the marketing says it does. Bank and email access are the two that matter.
- Set a spending limit or approval step for any transaction, so an agent cannot complete a payment unsupervised.
- Check who the data goes to. If the privacy notice lists partners you have never heard of, assume your data reaches them.
- Keep a manual route to cancel anything the agent signs you up for.
The wider pattern
This is the third regulatory signal on agentic AI in a fortnight, following a joint statement last month and continued discussions with developers. The direction is consistent: the UK is not banning the category, it is demanding that firms arrive with answers.
For context on how quickly the underlying models have changed, our piece on how open-weight models closed the gap explains the capability side, and the Ofcom review of smart speaker data retention covers the last time a voice product's data handling landed in front of a regulator.
The bottom line
The letter is a warning shot with a deadline attached: engage before launch, or expect the regulators to use the powers they already have. If you are buying or connecting an agent in the next few months, treat the permission screen as the important document, because it is the only part of the arrangement you control. Our guides to securing your accounts with passkeys and checking whether a site is safe cover the parts of this you can fix yourself today.

