TL;DR
The Financial Conduct Authority (FCA) is asking the companies behind general-purpose chatbots to warn users when their models produce financial information or advice. Jessica Rusu, who leads data, information and intelligence at the regulator, made the request at a Bloomberg UK Policy Series event on 19 October. The regulator also published good and bad AI practice it found while testing firms this year.
Disclaimers at the point of use
Rusu’s case rests on two consumer figures. Among less-experienced investors between 18 and 40, four in five have turned to AI when investing, yet only 40% of consumers realise there is no recourse if a general chatbot’s advice turns out wrong. Rather than widen its own remit, the FCA wants the boundary made visible where the answer appears, through what she called “clear, proportionate and consistent disclaimers” from AI providers.
The speech also set out the FCA’s reply to the Mills Review, the July report on how AI could change retail finance by 2030. Six of its seven recommendations are going straight into live work. Where the review proposed a new public-interest tool for building financial capability, the FCA will instead build on IOSCO’s first TechSprint, run through the FCA’s AI Lab.
What good and poor practice looks like
The published examples group into three themes. On consumers, one firm maintains a continually updated set of signs of vulnerability so its chatbot can pass people who may need help to human advisers. Poor practice included building tools with no defined need, overstating what they can do and shutting out vulnerable customers by default.
On governance, Rusu said too many firms still spend too little, and that senior managers should answer personally for the AI they run. One firm put proposed uses before an independent ethics panel; others bolted governance on after deployment. On resilience, firms should plan for failure, keep fallbacks and understand their dependence on a single model. Rusu added that incident reporting “must extend to significant AI failures and near-misses”.
Looking forward
Two more items are coming. A joint FCA and Bank of England survey, due next week, finds AI in use at 86% of firms, and the FCA plans to work with industry and frontier labs on a framework for agentic finance covering agent identity and delegated authority. On 13 October Rusu told MPs the FCA wants to automate its own supervision wherever it can. In our view, the disclaimer request is the part most likely to reach ordinary consumers, but it depends on chatbot makers choosing to act, since the FCA is asking rather than requiring.