Trading Tech Review
Menu
Risk & Rules

What Regulators Say About Using AI to Invest

Quoted from the regulator's own factsheet, with the parts we could not read in it clearly marked as such.

By Trading Tech Review EditorialPublished 3 min read
The pale stone and glass façade of a modern institutional building photographed straight on across an empty plaza under flat overcast daylight.
Authorisation is a checkable fact about a firm, not a quality of a tool.Image: AI-generated for Trading Tech Review

Key takeaways

  1. ESMA warns that AI tools can generate advice that could be inaccurate or misleading and may result in significant financial losses.
  2. The same factsheet notes that in the EU, providing investment advice is a regulated service that requires authorisation.
  3. ESMA points investors to a public register of authorised firms, so whether a provider is authorised is a fact you can check.
  4. Whether a tool is useful and whether anyone is accountable for it are separate questions.
  5. We quote only what we could read in the document, and say plainly which passages we could not.
In this article · 6 sections

The short answer

In March 2025 the European Securities and Markets Authority published a factsheet for investors on using AI tools for investing. Its central warning:

"AI tools can generate advice that could be inaccurate or misleading and that may result in poor investment decisions and significant financial losses."

That is a regulator stating the failure mode directly. Not that the tools are frauds — that producing confident, wrong output is a normal thing for them to do, and that the cost of acting on it lands on you.

The same document makes a second point that is easy to skate past and more useful than it looks: in the EU, giving investment advice is a regulated activity.

Advice is a regulated service

ESMA's wording:

"In the European Union, the provision of investment advice is an investment service regulated under the markets in financial instruments directive, meaning that it requires authorisation."

And immediately after, a pointer rather than an exhortation:

"Access here the list of authorised firms."

What makes this worth knowing is that it converts a vague worry into a checkable fact. "Should I trust this?" is unanswerable in the abstract. "Is whoever operates this an authorised firm, and do they appear on the register?" has an answer, and you can go and find it in a few minutes.

What we could not read, and why we are telling you

Being straight about this matters on a page about regulation.

The ESMA factsheet is laid out as a designed document, and several of its bullet points are drawn in an embedded font that our text extraction could not decode. We know those passages exist. We can see where they sit. We could not read them reliably enough to quote.

Summaries of this warning circulate widely, including some confident-sounding sentences about what public AI tools do and do not owe investors. We are not reproducing them, because we could not verify them against the document, and a page about what a regulator said is the worst possible place to paraphrase from a search result. If we get those passages read, we will add them and say so.

What we do quote above was read in the document itself, and the extracted text is kept in the repository so the quotations can be checked without taking our word for it. The method is in our editorial policy.

What a regulator expects of firms

A separate strand matters when the tool sits inside an authorised business rather than being a public website. ESMA's 2024 guidance to firms using AI in investment services expects them to comply with MiFID II, including the obligation to act in the client's best interest — set out in what an AI assistant can and cannot do, along with the risk list ESMA attaches to it.

The distinction worth carrying away is not that one kind of tool is good and another bad. It is that these are different arrangements in terms of who is accountable when something goes wrong, and that difference is invisible from the quality of the interface.

Where copy trading and signals sit

There is a third regulatory thread, and it is the most specific of all. ESMA's 2023 supervisory briefing on copy trading addresses when following someone else's trades stops being information and becomes a service a firm needs authorisation for — turning, broadly, on whether execution happens without you acting each time. That is covered in how signals are generated, with the regulator's own caveat that the assessment is case-by-case.

What this means in practice

  • Check authorisation before you check features. It is faster, it is a matter of public record, and it tells you what recourse exists.
  • Separate the tool from the accountability. A genuinely useful tool can come with no accountability at all; an accountable firm can offer a mediocre one.
  • Treat "inaccurate or misleading" as the expected case, not the edge case. That is the posture the regulator's own wording implies, and it is a reasonable one for anything covered in what the tools do mechanically.

This page describes what regulators have published. It is not legal advice, and it does not tell you whether any particular product is lawful where you live — see the risk disclosure.

Sources and references

  1. Using Artificial Intelligence for Investing: What you should consider — European Securities and Markets Authority (ESMA)Regulator · retrieved 5 October 2026

More on risk & rules and the ideas in this article.