Regulation
The FCA's Mills Review lands with AI agents already live in ten brokers' client accounts, and no rulebook written for them
The regulator's landmark review into AI and retail financial services calls artificial intelligence a defining force in UK finance by 2030, but opts for outcomes-based supervision over prescriptive rules even as deployment has already outrun the framework meant to govern it.
The FCA's Mills Review, published under executive director Sheldon Mills, is the UK regulator's first serious attempt to set out how it intends to supervise AI in retail financial services rather than simply warn about it. Its headline conclusion is unambiguous: AI will be a defining force in UK retail finance by 2030. What is more notable is the approach it takes to get there. Rather than writing AI-specific rules, the review commits to an outcomes-based supervisory model built around seven priority recommendations, betting that existing principles like Consumer Duty can flex to cover agentic systems without a bespoke rulebook that risks being obsolete before it is finalised.
The gap the review itself documents is the real story for anyone running a trading desk or building broker infrastructure. At least ten brokers and platform vendors had already wired AI agents into live client accounts during the first half of 2026, executing trades, managing positions, or acting on instructions with meaningful autonomy, before any regulator had written rules aimed specifically at that practice. That is not a hypothetical compliance gap. It is ten live deployments' worth of real client money moving through systems the FCA is only now formally turning its attention to, which puts the review several months behind the market it is meant to govern rather than ahead of it.
The review's risk framing centres on four shifts it says autonomy introduces: faster fraud through deepfakes and synthetic identities, system-level fragility from a small number of firms' shared dependence on the same handful of AI providers, a possible shift in who actually controls the customer interface as agents and platforms intermediate more of the relationship than brokers do, and an erosion of accountability as more decisions get delegated to systems that cannot be held responsible the way a licensed individual can. On the consumer side, the numbers are the part worth sitting with. Roughly 26 percent of UK consumers already trust general-purpose tools like ChatGPT, Claude and Gemini for financial advice, an estimated 9 million UK adults are already using AI apps for financial guidance in some form, and only around 40 percent of them correctly understand what formal recourse protections do and do not apply to that advice. Mills's own framing captures the shift: as autonomy grows, the nature of regulatory risk changes, not just its scale.
The caveat for anyone building or operating in this space is that outcomes-based regulation is a bet on supervisory capacity, not a settled answer. It gives the FCA room to adapt as agentic systems evolve without locking in rules that age badly, but it also means firms get less certainty about where the exact lines sit until enforcement actions start drawing them in practice, and the AI-enabled supervisory model the FCA proposes to police this is itself unproven at scale. For a broker or platform already running agents in production, the sensible read is that the absence of a written rulebook is not the same as an invitation, given Consumer Duty and existing conduct obligations already apply in substance, and being first through the door on an enforcement action grounded in outcomes rather than a checklist is a materially worse position than being able to point to a specific rule you followed.
Read the original: Finance Magnates - The FCA's Mills Review lands with AI agents already live in ten brokers' client accounts, and no rulebook written for them. Commentary is the independent editorial view of Share Trading; the original article is credited to its publisher.