TL;DR

Jessica Rusu, who runs data, information and intelligence at the Financial Conduct Authority, told MPs on 13 October, according to the published transcript, that the regulator plans to hand as much supervision and enforcement work to AI as possible. Her words to the Commons Treasury Committee were “to automate wherever we can” and “to intervene at scale”. At the same session, chief executive Nikhil Rathi said he was not yet persuaded the FCA needed “lots more powers”, and it is not writing AI-specific rules for firms. The Financial Times reported the plans on 15 October.

According to Rusu, AI already handles unstructured text well, which speeds up case officers’ analysis, and the FCA has been looking at embedding it in authorisation, supervision and enforcement. At this point, she said, a case officer or other human stays involved in every decision.

Agents on surveillance data

One of the first jobs for the FCA’s agentic tools, Rathi said, has been cleaning the billion rows of market surveillance data it receives each week. Where a submission fails to reconcile, the tool can go back to the firm, so that staff reach the highest-risk issues sooner. Rusu also wants to “shift left”, catching harms at the authorisation gateway.

The direction comes from the Mills Review, published in July by Sheldon Mills, who had been an FCA executive director. Among its recommendations were an AI-enabled, agentic approach to supervision and greater powers for the regulator. It also asked whether ChatGPT, Claude, Gemini and similar models should be brought under financial rules. The FT reports that one in five UK adults surveyed for the review were willing to let an AI system take money decisions on their behalf.

Where the perimeter sits

Rathi told the committee that a chatbot offering general information, with no personal recommendation that ends in a transaction, is not wholly inside the regulated perimeter. That line is “a very live discussion”, he said, and regulating large language models as such would be a question for government and Parliament.

Looking forward

A formal FCA response to the Mills Review should come from Rusu in a speech expected in the week after the FT’s report, along with what staff learned from visiting firms about good and bad AI practice. The regulator’s AI Live Testing service launched in December 2025; its first cohort included Monzo, NatWest, Scottish Widows and Santander. In our view, the change UK firms will feel first is not a rulebook but software reading their submissions and raising discrepancies on its own. Start-ups building consumer AI money assistants should follow the perimeter debate, which Rathi has left open.