Regulation / Jul 6, 2026 / 4 min
Britain Found Its Adviser. It Isn't on the FCA Register.
On July 6, the FCA's Mills Review found 26% of UK consumers already trust ChatGPT, Claude, and Gemini for financial advice — and 11 million adults are open to AI making money decisions for them — while every one of those tools sits outside Britain's regulated perimeter, forcing the watchdog to decide within six months whether to police frontier models or watch the advice gap fill itself with unprotected chatbots.
The UK's financial watchdog just published the first regulator-led global review of AI in retail finance — and its central finding is that millions of Britons are already taking money advice from models the FCA cannot touch.
The numbers:
- 26% of UK consumers trust general-purpose tools such as ChatGPT, Claude, or Gemini for financial advice — with limited awareness that formal redress routes do not apply (FCA Mills Review, July 6, 2026)
- 1 in 5 UK adults — roughly 11 million people — are already open to AI making financial decisions for them, with strongest demand in debt, pensions, and investments (Finextra)
- Only 9% of consumers use traditional regulated advice annually — while £300 billion sits in low-interest accounts the review says AI could help mobilize (FCA Mills Review)
- 75% of UK financial services firms now use AI; 81% globally are adopting it at some level (FCA Mills Review; Reuters)
Why July 6 matters:
Executive Director Sheldon Mills delivered the Mills Review to the FCA board on Monday — the first review of its kind commissioned by a financial regulator anywhere, drawing on 140 written submissions and a Yonder Consulting survey of more than 5,000 UK consumers.
Mills told Reuters Britain should decide within three to six months whether to "secure and adapt" the regulatory perimeter so general-purpose LLMs fall inside it. The FCA's own consumer guidance already warns that ChatGPT-style tools are "not regulated by us in the way that regulated financial advice" is — meaning no Financial Ombudsman Service, no Financial Services Compensation Scheme (FCA InvestSmart).
The autonomy spectrum:
The review maps how consumer finance is shifting along five stages — from AI as operator's tool to AI as autonomous observer acting within preset limits. Mills writes that as systems move rightward on that spectrum, "accountability becomes harder to trace, especially where models change over time and firms depend more heavily on model providers."
Demand is already at stage four. The gap is governance at stage zero.
What regulators elsewhere are doing:
- Singapore (July 3): MAS published SAFR — runtime governance checkpoints that verify agent actions before execution
- Threadneedle Street (June 30): Bank of England Deputy Governor Sarah Breeden floated market-wide kill switches if faulty AI agents trigger meltdowns
- London (July 6): Mills proposes perimeter expansion, a public-interest AI financial-capability service, and an "agentic supervisory model" so the FCA can monitor system-wide correlated failures across shared models and cloud providers
Three capitals. Three philosophies. All racing the same agent deployment curve.
The concentration risk:
Mills warns that shared reliance on the same frontier models, cloud stacks, and hyperscalers could produce "correlated behaviour, herding, opacity and common points of failure" across the financial system. A model outage or security breach at OpenAI, Anthropic, or Google does not stay inside one bank's risk committee — it propagates through every firm plugged into the same API.
OpenAI began letting U.S. users connect financial accounts to ChatGPT in May 2026. The Mills Review notes these integrations are exactly how general-purpose tools start shaping regulated outcomes from outside the perimeter.
Seven recommendations, one deadline:
- Secure and adapt the regulatory perimeter (3–6 month review)
- Strengthen system-wide coordination across UK authorities
- Monitor the transition to autonomous models
- Scale the FCA's AI Lab
- Enable foundations for agentic finance
- Build an AI-enabled agentic supervisory model
- Develop a trusted public-interest financial-capability service
FCA Chair Ashley Alder said the watchdog must "keep pace with a rapidly changing environment" while holding to its principles-based Consumer Duty framework (Finextra).
Mills himself is blunt in the foreword: "The question is no longer whether to allow AI, it is who is AI going to serve?" His answer: citizens — not platforms capturing the customer interface.
Convina's view: The Mills Review is the most honest admission yet that activity-based regulation broke the moment ChatGPT started answering pension questions. London is not rushing to ban frontier models — it is trying to decide whether OpenAI and Anthropic are infrastructure providers or shadow advisers, and the six-month perimeter review is the fork. Singapore is building runtime guardrails. The Bank of England wants kill switches. Britain wants to redraw the map. All three are behind the consumer: 26% already trust the chatbot, and the FCA's own data says 11 million adults want an agent to act for them. The advice gap did not disappear — it relocated to a jurisdiction the FCA does not supervise, and the Mills Review is the first official document to say that out loud.