The FCA-commissioned Mills Review marks a significant regulatory shift, moving away from a hands-off approach to AI agents making financial decisions, says Omar Salem
Pablo Neruda, the Chilean Nobel laureate for literature, said “you can cut all the flowers but you cannot keep spring from coming”. The FCA has finally accepted this for the regulation of AI in financial services.
The FCA-commissioned Mills Review of AI in retail financial services marks a significant shift in tone. Until now, the FCA position was that no new AI rules were needed. Existing frameworks, applied in an outcomes-focused way, would do the job.
The FCA’s instinct was understandable. The UK’s principles-based regime is one of its strengths. It gives firms room to innovate and regulators room to adapt. But that position was always going to collide with reality given the enormity of change coming from AI. In the meantime, the FCA pushed the burden onto City firms to work out for themselves how the FCA handbook applied to a new transformative technology.
Now the Mills review, authored by the architect of the FCA’s consumer duty, has landed. It goes much further than “no new AI regulation” and makes a raft of recommendations. The review is not just about chatbots, it is about how much AI should be able to do for you and what protections you should have. Perhaps most of all, it is about AI agents.
AI agents can do more than answer questions. They can break down a task, use tools, access documents on your computer, and act on your behalf with third parties like retailers, banks and wealth managers. In financial services, that could mean an AI agent comparing savings accounts, switching insurance, monitoring pension contributions, consolidating pots, managing investments, initiating payments, or managing debts. The idea is that the consumer sets the goal and boundaries and the agent does the legwork.
Can you trust agentic AI? #
This could be powerful. The review points to long-standing weaknesses in retail finance: only nine per cent of consumers use traditional advice, around 900,000 people are unbanked, and £300bn sits in low-interest accounts. AI agents could help consumers who are overwhelmed, disengaged or poorly served.
But spring brings weeds as well as flowers. Like many others, I wonder how much I could trust agentic AI. It certainly feels a bit uncomfortable to this millennial.
The review found that one in five UK adults are already open to AI making decisions for them, while around 26 per cent trust general-purpose tools such as ChatGPT, Claude or Gemini for financial advice. These figures are likely to change as AI use becomes more prevalent.
Yet there are lots of questions for today. If I give an agent permission to make payments for me, what happens if it empties my bank account? If an agent invests funds badly, who is responsible? If firms use AI to personalise prices and journeys, does that mean better products and prices, or unfair discrimination?
This is why, whereas the FCA previously left firms to decide how existing requirements applied to their use of AI, Mills now says the FCA should consider mandating requirements for AI agents.
But that is not all. The review also recommends that the FCA use agents to supervise firms and their agents. That may sound a little circular but it makes sense that the FCA uses the best available technology. AI will guard the guardians.
The practical message for City firms is clear. As the review states, strong AI governance is becoming a core capability. AI will touch every aspect of most City firms and will require an across-the-board view of its commercial, risk and regulatory impact. The era of working out AI compliance in something of a regulatory void is ending. There should be more clarity, but also more scrutiny.
The FCA has not abandoned its principles-based model. Nor should it. But the Mills Review sets out how AI is moving from assistance to action. The FCA must keep pace.
Omar Salem is a financial regulation and technology partner consultant at Pierson Ferdinand UK LLP