Anyone who has been through a network file check knows how it works. Someone picks a sample, maybe ten cases, maybe twenty, and reads them properly. The fact find, the research, the suitability letter, the file notes. If those hold up, you get a clean report and everyone goes back to work. Most of the cases you wrote that quarter are never read by anyone at all.
I’ve been thinking about that since reading the FT’s report last week, because the FCA wants to change the maths. The regulator plans to become what it calls an “agentic regulator”, using AI to automate much of its supervision and enforcement work. Jessica Rusu, its chief data, information and intelligence officer, told MPs on the Treasury Committee that the FCA already uses AI in ways its officials oversee and direct, but the aim is to go further, “to automate wherever we can … and to intervene at scale”.
“Agentic” is a word you’ll hear a lot more of. It just means AI that does things rather than only answering questions. It reads, decides what matters and acts, within limits someone has set for it. Building that kind of supervisor was one of seven recommendations in the Mills Review, which the FCA published in July. Sheldon Mills, then an executive director at the regulator, told it to “build and adopt an AI-enabled agentic supervisory model”. The FCA appears to have taken that one to heart.
So what would it actually read? Rusu told the committee that AI was particularly effective at analysing and summarising the “unstructured text” firms send the regulator in emails, documents and interviews. Unstructured text is a dull phrase for something mortgage firms produce by the tonne. Suitability letters, file notes, complaint responses, the free-text boxes on a fact find. Nearly all of it written by people, for people, on the quiet assumption that a human would only ever read a sample.
A machine doesn’t need a sample. And the thing it’s likely to be good at is exactly what human reviewers are worst at, which is spotting patterns across thousands of documents. The same reason for recommendation pasted into forty different files. A cluster of complaints about one product that only shows up when you look at a whole year. Suitability letters that read beautifully but never seem to mention the cheaper option. A file checker looking at ten cases would never see any of that. I’m speculating here, because the FCA hasn’t said how it will use the technology. But if I were designing an AI supervisor, that’s where I’d start.
What the FCA isn’t doing is just as interesting. The same FT report says the regulator is stopping short of imposing new rules on the sector. Nikhil Rathi, the chief executive, told the committee his thinking on AI “has been evolving”, but he wasn’t convinced the FCA needed “lots more powers until we have really demonstrated that we have used our powers to the fullest extent”.
I think he’s right, and I suspect most firms will be quietly relieved. The way I read it, there’s no AI rulebook coming. If you use AI to help draft a suitability letter, there’s no special AI test to pass. The test is the one you already know. Was the advice suitable, and was the outcome good for the customer? Consumer Duty doesn’t care whether a person or a model wrote the words.
Then there’s the bit of Rathi’s evidence I keep coming back to. He said he’s often asked whether the FCA should regulate AI models themselves. His answer, as the FT reported it, was that where a model “is providing information but is not providing an individualised recommendation which is leading to a transaction by way of business, it does not fall fully within the regulated perimeter”. Then he added, “But I think that is going to be a very live discussion: where do you draw that boundary?”
Research for the Mills Review found that one in five UK adults are already open to AI making financial decisions for them. Some of your clients are already asking ChatGPT which lender to try, how much they can borrow, and whether to fix for two years or five. As long as the answer counts as general information, nobody regulates it. If it’s wrong, there’s no complaints route and no compensation scheme standing behind it. Then the client rings you, and from that moment the regulated responsibility is yours.
Put those two things side by side and you get an odd picture. The AI inside regulated firms is about to be watched more closely than any adviser has ever been, by a regulator using AI to do the watching. The AI on your client’s phone, the one they consulted before they called you, mostly sits outside the fence.
I don’t think that’s a scandal. Regulating every chatbot that mentions a mortgage would be close to impossible, and Rathi is right that the boundary is genuinely hard to draw. But it does mean the regulated adviser becomes the place where the risk lands, whichever machine created it.
There’s more to come in the next few days. Rusu is due to set out the FCA’s full response to the Mills Review in a speech, and she told MPs the regulator has been visiting firms to see how they’re using AI in practice. It has, in her words, “learnt a lot about good and poor practices”, and says it will publish what it found. I’ll be reading that closely. If the FCA has seen poor practice in AI-assisted advice, I’d much rather find out what it looks like from a published report than from an AI supervisor working through someone’s files.
One last thought. If brokers use AI to write suitability letters, and the regulator uses AI to read them, the letter becomes a conversation between two machines. If we’re honest, the client doesn’t always read it properly either.
So where in that chain does the human judgement actually sit, and could you point to it if a machine asked?
Sources:
Financial Times, “FCA plans to embrace AI to automate financial regulation”, Martin Arnold and Emma Dunkley, 15 October 2026: https://www.ft.com/content/3a9364ec-6930-4877-a1a2-c053dd689760
Retail Banker International, “UK’s FCA seeks to automate oversight via AI – report”, 16 October 2026: https://www.retailbankerinternational.com/news/uk-fca-automate-ai-oversight/
FCA, “AI and the future of retail financial services (The Mills Review)”: https://www.fca.org.uk/publications/corporate-documents/mills-review