The UK's Financial Conduct Authority is calling for expanded powers to police the use of AI in financial services, warning that regulators are in an "arms race" with the technology as one in five UK adults say they are open to letting an AI model make personal money decisions for them. Sheldon Mills, an executive director at the FCA, set out the case in a commissioned report published Monday, which recommends a review in the next three to six months into whether consumer chatbots like ChatGPT, Claude and Gemini should sit inside the regulatory perimeter. The regulator's board will discuss the findings before responding.
Mills told the Financial Times the FCA itself would need to adopt AI to keep pace with the "speed, pace, and scale of change" reshaping the sector. The report frames the technology as double-edged: hyper-personalization could match products to individual needs, but could equally enable bias, opaque pricing and what Mills calls "personalized manipulation." His research found that a fifth of UK adults would already trust an AI model with savings or borrowing decisions, despite there being no compensation route if the advice goes wrong.
The regulatory gap Mills is targeting is specific. Authorized firms in the UK face "reasonably strict" rules when they make product recommendations to consumers. A large language model that responds to a prompt about a mortgage or an ISA faces none of them, even when the output is functionally similar. Mills wants the FCA to determine whether a conversational response to a financial prompt counts as "recommendation, or guidance" under existing law — a question with real consequences for OpenAI, Anthropic and Google, whose consumer products are already being used this way.
Key facts
- 01One fifth of UK adults are open to using AI models like ChatGPT, Claude or Gemini to make personal financial decisions, per FCA-commissioned research.
- 02FCA executive director Sheldon Mills wants a review in the next three to six months into consumer harm from AI-driven finance advice.
- 03The report recommends expanding FCA powers over critical third parties including Anthropic, OpenAI, Amazon, Google and Microsoft.
- 04Mills is leaving after eight years at the FCA, having also greenlit a 12-week contract with Palantir to test AI for financial crime detection.
- 05AI could extend advice usually reserved for clients with £10mn in assets to people earning £20,000 a year, Mills argues.
The upside case in the report is access. Mills argues AI could "democratize" finance by widening the pool of consumers who can get sophisticated advice, extending services traditionally reserved for clients with £10mn in savings or assets to people earning £20,000 a year. He wants the FCA to convene public and private sector groups to build a free "AI-enabled financial capability service" for the British public, positioning the regulator as an active participant rather than a bystander.
The FCA also wants leverage over the infrastructure underneath. The report recommends boosting the regulator's authority under the "critical third parties" regime, which lets it supervise key technology providers to the financial sector — a list that would include Anthropic, OpenAI, Amazon, Google and Microsoft. Designation would allow the FCA to require annual self-assessments and scenario testing for severe disruptions. The UK government has not yet decided which Big Tech firms will be formally designated.
“Some firms have said to us that they feel that this could be an economically equivalent type of service that isn't regulated [and] sits outside of the regulatory perimeter”— Sheldon Mills, Executive Director, Financial Conduct Authority
Beyond the critical third parties framework, Mills flags the "designated activities regime," which lets the FCA regulate specific activities without requiring the underlying firms to seek full authorization. That would give the regulator a more surgical tool: instead of forcing a general-purpose chatbot to become an authorized financial adviser, it could regulate the specific act of giving financial guidance through such a tool.
The report also treats AI as a security problem. Deepfakes, synthetic identities and personalized social engineering are "taking fraud and cyber risks into a new era," it says, and Mills argues the same technology has to be used to defend the financial system as attackers scale up their tactics. That framing is why the FCA sees itself as needing to run at the same pace as the labs it may soon supervise.
Financial services firms are already piloting AI agents that can autonomously carry out transactions for companies and consumers, moving past chat and into execution. Mills, who is leaving after eight years at the FCA, is clear that accountability cannot follow the model.
“You need a human on the hook for what they're doing”— Sheldon Mills, Executive Director, Financial Conduct Authority
That principle — a named human liable for what an autonomous system does — is likely to become a central design constraint for any bank or fintech deploying agents in the UK. It is also a marker of how the FCA plans to draw the line between AI as a tool and AI as an actor.
The regulator itself is not immune to political scrutiny over its own AI choices. The FCA signed a 12-week contract with Palantir to test whether the US company's systems can help detect financial crime, drawing criticism from MPs who worry the deal could route sensitive UK financial data to US authorities. The FCA and Palantir have denied that is the case, and Mills declined to comment on the contract.
For the AI industry, the FCA's move is the first serious signal from a G7 financial regulator that consumer chatbots may be pulled inside the perimeter for financial guidance specifically, rather than through generic AI legislation. If the review concludes that a prompt-and-response exchange about savings is "economically equivalent" to regulated advice, the compliance surface for OpenAI, Anthropic and Google in the UK gets materially larger — and the template will travel. Watch whether the FCA opts for the designated activities regime, which pins new rules on the act rather than the firm; that would be the fastest route to regulating chatbot advice without waiting for primary legislation, and the one that changes product design decisions in San Francisco this year rather than next.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
