Does your chatbot say it is an AI? Eight of twelve do.

This article covers what the rule says, what we observed, and how we have chosen to handle it ourselves. It is not legal advice, and it deliberately avoids the question of who is or is not compliant.

What does the rule actually say?

Article 50(1) of the AI Act (Regulation (EU) 2024/1689) requires providers to ensure that people interacting directly with an AI system are informed that they are doing so. There is one notable exemption: the duty does not apply where it is obvious to a reasonably well-informed and attentive person, given the circumstances, that they are talking to an AI.

That exemption matters. A company presenting a clearly robotic button-menu may well fall within it. A fluent, humanlike assistant with a human-sounding name is a harder case.

Whether any individual company falls inside or outside the exemption is an assessment for supervisors and ultimately courts, not for us.

What did we observe?

As part of our CRIX measurement of Swedish consumer lending chats, we recorded what each of the twelve publicly reachable assistants displayed about its own nature. We deliberately waited three weeks after the requirement took effect, and then observed every chat in a single sweep, so that each company had a reasonable opportunity to bring its disclosure into line before we looked.

Eight of the twelve state that the customer is talking to an AI. Four of those also warn that the answers may be wrong. Three present the assistant as something other than an AI. One displayed nothing at all.

Why does this matter in practice, not just on paper?

The caution is warranted. In the same measurement, 19 per cent of the answers given contained something untrue.

Ordinary questions about co-borrowers, income and payment defaults, answered fluently, confidently and incorrectly.

This finding makes informing consumers that they are talking to an AI more than a regulatory formality. It is the one signal that tells them how to read the answer, and to treat it with care. Banking is built on trust: a consumer who interacts with a bank has every reason to trust what they are told, and will typically act on it. That is exactly why they need to know when the answer comes from a machine that can be wrong.

As long as answers can be wrong, telling people what they are talking to is part of answering honestly.

So the four companies that both state it is an AI and warn that answers may be wrong are not being over-cautious, they are simply doing what they should. The disclosure and the error rate belong together.

Who supervises this in Sweden?

Despite the rule applying since 2 August 2026, Sweden has not yet settled who will supervise it. Interim authorities have been appointed, among them Finansinspektionen, IMY and PTS, but no authority has yet been designated for the chatbot transparency duty specifically.

Waiting for that to resolve is not much of a plan. The rule applies today, and avoiding a position on it is not primarily a legal risk. It leaves the company’s own customers without the one warning that would make them careful.

How we handle it ourselves

We are not a compliance advisor, so we will simply describe our own reasoning. Bill, our assistant, states that it is an AI, warns that answers can contain errors, and shows which sources each answer is built on. We do this not because a rule forces us to, but because the disclosure and the accuracy work belong together: telling a person they are talking to a machine is what makes them apply the healthy caution that even a well-built system deserves. Transparency is not a legal tax on AI, it is part of what makes AI usable in a regulated industry built on the foundation of trust.