A disclaimer does not move the line between chat and advice

A label is not a rule

A disclaimer is a sticker. It records what the manufacturer says the product is. It does not change what the product does.

The Financial Conduct Authority — the body that decides who in Britain may be paid to tell people what to do with their money — is reported to be calling on artificial intelligence providers to attach disclaimers to services that give consumers financial advice through agentic systems. The phrasing of that call contains its own refutation. If the service provides advice, a notice stuck on the front saying it does not is contradicted by the thing it is stuck to.

British law follows the conduct, not the branding

The regulatory perimeter in the United Kingdom is drawn around activities. Giving a personal recommendation to a consumer about what to do with an investment is a regulated activity, and carrying it on by way of business without authorisation is not permitted. That test looks at what is being done and for whom. It has never been satisfied or escaped by what the provider calls itself.

Firms will reply that there is a real and long-standing distinction between generic information and a personal recommendation, and that most of what a chatbot produces sits on the harmless side of it. That is true of general-purpose assistants answering general-purpose questions. It is much less true of the systems actually being built. An agentic financial assistant is personalised by construction: it reads the balances, the holdings, the income, the stated goal, and produces an answer shaped to that person. Personalisation is precisely the ingredient the perimeter test is looking for.

The word doing the work is “agentic”

Agentic means the software acts. It does not stop at a suggestion; it opens, switches, moves, cancels, subscribes. That is the whole commercial point, and it is why these products are being built rather than another chat window.

So a disclaimer about advice is a warning about the wrong risk. Advice risk is the risk of a bad recommendation that a human then chooses to follow, with a pause in between where the consumer can think. Agentic risk is the risk that the pause is gone — that a misread instruction, a hallucinated product feature or a poorly specified goal results in money having already moved. You cannot discharge that with a line of small print displayed before the agent is given permission to act. The small print will simply become the permission.

Disclosure already had its decade

The British regulator spent years establishing that warnings on their own rarely change behaviour, and then built the Consumer Duty around a different proposition: firms must deliver good outcomes for retail customers, not merely describe bad ones in advance. That was a deliberate shift away from disclosure as the primary remedy. Reaching for disclaimers in 2026, for the most consequential new distribution channel in consumer finance, is reaching for the instrument the regulator itself concluded was insufficient.

The strongest case the other way

The case for disclaimers is speed and reach, and it is not a weak one.

Many of the firms building these agents are not authorised financial firms. Some are not in the United Kingdom at all. The FCA cannot supervise a model developer in California the way it supervises a bank in Canary Wharf, and formal perimeter guidance takes consultation, drafting and time the market will not wait for. A public expectation that these services carry honest labelling is available immediately, costs nothing, and creates a documented standard that makes later enforcement easier: a provider that ignored a stated expectation looks very different in front of a tribunal from one that never heard it.

There is also a genuine consumer benefit in calibration. People trust fluent text. Being told plainly that the thing talking to you is not a regulated adviser and owes you no duty is better than not being told.

All of that justifies a disclaimer as an interim measure. None of it justifies a disclaimer as the answer. And the reach problem has a solution the regulator already possesses: wherever an agent actually touches a British consumer’s money, there is an authorised firm in the chain — the bank holding the account, the platform executing the switch, the provider accepting the application. The grip is there. It just has to be used.

What would settle it

Two documents would settle it. First, perimeter guidance that says when the output of an agentic system counts as a personal recommendation, with worked examples, so builders know which side of the line they are on before they ship rather than after a complaint. Second, a rule that where an agentic service initiates a transaction on a retail customer’s account, a named authorised firm is accountable for the outcome, whoever wrote the model.

Until those exist, a disclaimer is not consumer protection. It is a receipt for the transfer of risk from the firm that built the agent to the person who used it.