WallStreetDebug>_ Wall Street, debugged.
Markets
Subscribe
Regulator watch

The FCA wants ChatGPT, Gemini and Claude to warn users about money advice. Here is what the speech actually says

A UK regulator has asked the companies behind general-purpose chatbots to warn people at the moment AI gives them financial information. It is a request, not a rule, and the details matter.

By WallStreetDebug DeskPublished 4 min readNot financial advice
4 in 51Less experienced investors aged 18 to 40 who have used AI to help with investing
40%1Consumers who correctly recognize there is no recourse after acting on general-purpose AI advice
6 of 7Mills Review recommendations the FCA is embedding in live work

On Monday, October 19, 2026, Jessica Rusu, the UK Financial Conduct Authority's chief data, information and intelligence officer, spoke at Bloomberg's UK Policy Series. Headlines focused on one line. We read the FCA's own published text of the speech, plus the good and poor practice guidance it released the same day, to separate what was asked from what was not.

The ask

The speech says consumers are turning to general-purpose large language models, naming ChatGPT, Gemini and Claude, for financial information and advice, while regulated firms begin to integrate their services into AI platforms. That can widen access, Rusu said, but it can also blur the boundary between regulated and unregulated activity and the protections that come with it.

The FCA's stated answer is not to expand its remit but to make the existing boundary clearer. In its words: "we are calling on AI providers to help consumers understand the limits of financial information and advice generated by their models through clear, proportionate and consistent disclaimers." The speech adds that warnings may work better at the point where AI-generated financial information is delivered, not on a regulator's website.

The two numbers behind it

Rusu tied the request to two consumer figures. Four in five less experienced investors aged 18 to 40 have used AI to help with investing. Yet only 40% of consumers correctly recognize that there is no recourse if something goes wrong after acting on advice from a general-purpose AI tool.1

The speech does not give the sample size, the survey date or a margin of error for either figure, so treat them as the regulator's headline readings, not a dataset we can check. The gap between the two is the point: heavy use alongside a widespread belief that someone would be accountable.

What else was in the speech

  • The FCA says it is embedding 6 of the 7 recommendations from the Mills Review, published in July, into live work. Instead of building a new public financial-capability tool, it will build on IOSCO's first TechSprint, run with its AI Lab, to help investors spot scams, verify information and challenge unreliable AI outputs.
  • A joint survey with the Bank of England, to be published next week, shows 86% of firms now use AI, according to the speech. We have not seen the survey itself.2
  • Existing incident reporting should extend to significant AI failures and near-misses. Rusu called near-miss reporting for agentic AI, meaning systems that act for a user, a regulatory priority.
  • The FCA will work with industry, other regulators and frontier labs on a framework for agentic finance covering agent identity, delegated authority, accountability and execution. The speech gives no timetable.

Good and poor practice, same day

Alongside the speech the FCA published a guidance page on AI in financial services. It states that the examples do not create new requirements and that existing rules apply. Firms that use AI responsibly, it says, define the consumer need first, test outcomes across different consumer characteristics, tell consumers how they used AI and its limits, and monitor live outcomes. It also describes choosing where AI should support, automate or abstain, redirecting or escalating to a person in some cases.

The speech lists poor practice as well: building AI tools without a clearly defined need, excluding vulnerable consumers by default, overstating capabilities, deploying automated customer journeys without enough testing, and retrofitting governance after deployment. On resilience it flags reliance on a single model.

Why this connects to testing

A disclaimer is only useful if the underlying behavior is understood. Our own coverage this month of AI answers to money questions, including a study of how often chatbots got money answers wrong, shows why regulators are looking at the boundary. A same-day arXiv audit, broken down here, found that seven models moved their advised stock share by 24 to 35 points when only a client's risk answer changed. A warning label does not fix that, but it tells the user the output is not an assessment of them.3

What it shows / What it doesn't

It shows

  • That the FCA is publicly asking AI providers for clear, proportionate and consistent disclaimers on financial information and advice.
  • That the regulator frames this as clarifying the existing boundary, not widening its remit, and sets no deadline in the speech.
  • That the FCA cites 4 in 5 younger, less experienced investors using AI, against 40% understanding there is no recourse.1

It doesn't show

  • That any disclaimer rule exists. The speech is a request, and the regulator says the examples in its guidance create no new requirements.
  • How the consumer figures were measured. The speech gives no sample, date or margin.
  • Whether chatbot financial answers are accurate. That is a separate question we test elsewhere.
  • What the Bank of England survey says. It had not been published when we wrote this.

Why it matters

If you use a chatbot to think through money, the regulator is saying out loud what the regulated world assumes: there is no adviser relationship and probably no recourse. Whether AI providers adopt consistent warnings is now an open question to watch. Until then, the habit that helps is simple: ask the model what it assumed about you, and check anything that matters against a primary source. This is information, not advice.

Data notes

Each small number in the text points to a note below: the sample size, the period, the source and, when the source publishes one, the margin of error. "Not disclosed" means the source does not give it.

  1. n: not disclosed · consumer research cited in the speech, date not disclosed · FCA (Rusu speech, Oct 19, 2026) · margin: not disclosed
  2. n: not disclosed · joint FCA and Bank of England survey, publication due the week of Oct 26, 2026 · FCA (Rusu speech, Oct 19, 2026) · margin: not disclosed
  3. n=800 households, 7 models · SCF 2007-09 panel, advice generated 2026 · Bakshi et al., arXiv:2610.23060 · 95% bootstrap intervals

Sources

"Figures checked" means we opened the source and compared the figures we quote with it. It confirms what the source says, not that its claims are independently true. How we label sources

How we used AI: We used AI to draft this summary. Every quote and number was then checked against the FCA's published speech and guidance page, opened on Oct 19, 2026, in a separate pass by the same system.

Not financial advice. For information and education only. AI model outputs and research results can be wrong and do not predict future returns. Nothing here is a recommendation to buy or sell any asset. See our disclaimer and methodology.

RegulationLLM adviceConsumer protectionFCA

Suggest a tool or study

Know an AI tool, a study or a claim we should look at next? Email us at [email protected].

Related