The FCA's landmark Mills Review lands with a warning that regulators are locked in an "arms race" with the technology, and a recommendation that could pull general-purpose AI models into the perimeter of financial regulation for the first time.
Britain's financial regulator is asking a question that would have sounded far-fetched a year ago: should the large language models that millions of people now consult about their savings be treated as regulated financial tools?
On Monday, the Financial Conduct Authority published the Mills Review, a report the regulator describes as the first of its kind commissioned by any financial watchdog in the world. Its central recommendation is blunt. Within the next three to six months, the FCA should decide whether general-purpose models such as OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini belong inside the regulatory perimeter when they start steering people's financial choices.
Sheldon Mills, the FCA executive director who led the review, framed the challenge in stark terms. He called the gap between how fast AI is being adopted and how fast regulators can respond an "arms race."
The Consumer Habit That Worries Regulators
The number driving this is hard to ignore. Research commissioned for the review found that a fifth of UK adults, roughly 11 million people, are open to letting AI act autonomously on their finances within goals they set in advance. Many already lean on chatbots for guidance on saving and borrowing.
The catch is what those people don't know. Consumer protections that apply to regulated financial advice do not extend to a conversation with a general-purpose chatbot. Someone asking Claude whether to move their pension is not covered the way they would be if they walked into an adviser's office, yet the review found limited public awareness of that distinction.
Mills put his finger on the specific tension. He questioned whether these conversational systems are quietly crossing a line, becoming something closer to a recommendation or formal guidance, the kind of activity that already triggers strict rules when a regulated firm does it. A human adviser giving equivalent guidance faces obligations a chatbot simply doesn't.
Seven Recommendations, One Big Idea
The review sets out seven recommendations for the FCA board. The one drawing the most attention is the first: "secure and adapt" the regulatory perimeter by studying the scale, nature, and impact of general-purpose models that currently operate outside it.
The other six range across strengthening system-wide oversight, monitoring the shift toward autonomous models, scaling up the FCA's internal AI Lab, laying the groundwork for what the report calls agentic finance, and building an AI-powered supervisory model so the regulator can watch the market using the same class of tools the market is deploying. There's also a proposal for a public-interest service that would offer free, AI-enabled financial guidance to people who can't afford a traditional adviser.
That last idea points to something the FCA is careful to stress. Mills sees a genuine upside here. AI could widen access to financial advice that has long been the preserve of wealthier clients, and the regulator has no interest in banning it.
Accountability Cannot Be Automated Away
Where Mills draws a firm line is on responsibility. As firms roll out AI systems that can act with little human intervention, the review insists someone must still answer for what those systems do.
"You need a human on the hook for what they're doing," Mills said.
That principle sits at the heart of the FCA's existing approach, which leans on the Consumer Duty and the Senior Managers and Certification Regime rather than a bespoke AI rulebook. The regulator has said repeatedly it does not plan to introduce AI-specific regulation, preferring to stretch its principles-based framework to cover new behaviour. Whether that framework can hold as AI agents begin comparing products and making choices on people's behalf is precisely what the review is testing.
The Concentration Problem
Beyond consumers, the review flags a structural danger that could reach across the whole financial system.
A recent survey cited in the report found 81% of financial firms globally are adopting AI at some level, with 40% already scaling it or using it to transform how they operate. In Britain, most of that activity still sits in lower-risk back-office work. But firms are increasingly pushing AI into customer-facing roles, handling complaints and offering investment guidance.
The problem is where all these firms are getting their AI. If banks, insurers, and investment houses all depend on the same handful of model providers, cloud platforms, and infrastructure, the system starts to share a single set of weak points. The review warned that this shared reliance could produce correlated behaviour, herding, and common points of failure. When everyone runs on the same model, everyone can fail the same way at the same time.
To address that, the report recommends expanding the FCA's oversight of major technology providers through the UK's critical third parties regime. That mechanism could eventually cover the biggest names in the sector, including OpenAI, Anthropic, Amazon, Google, and Microsoft. The government has not yet decided which providers will be formally designated, a step that would bring greater resilience testing and disclosure requirements.
A Global Backdrop of Divergence
The FCA is not acting in isolation, and its chosen path stands apart from others.
The context for the review includes the operational and cyber risks tied to frontier systems like Anthropic's Mythos-tier models, alongside the harder-to-predict behaviour of agentic AI that acts on its own. Regulators everywhere are wrestling with the same forces.
They are not answering the same way. The European Union's AI Act classifies financial AI applications as high-risk and imposes strict, prescriptive requirements. Britain, outside the EU, is charting a different course that Mills has described as needing to stay agile and proportionate. Ashley Alder, the FCA's chair, reinforced that stance, saying the regulator has to keep pace with a fast-changing environment while sticking to the principles-based, outcomes-focused approach it has taken so far.
Industry Wants Rules, Just Not Heavy Ones
The financial sector's response has been supportive with a clear caveat.
UK Finance, the trade body for banking and financial services, acknowledged the need for clear rules while cautioning against anything so prescriptive it chokes off innovation. Its position is that a principles-based approach lets firms build new products without leaving consumers exposed.
Brian Byrnes, director of personal finance at savings app Moneybox, went further in backing the direction of travel. He argued that if AI is influencing financial decisions or providing guidance, it should meet the same standards and consumer protections as any other regulated provider. Innovation and protection, in his view, are not rival goals that force a trade-off.
OpenAI, Anthropic, and Google were not immediately available for comment when the review landed. The FCA board will now weigh the seven recommendations before setting out its formal response, with the three-to-six-month clock on the perimeter question already ticking.
Comments 0
Join the discussion and share your perspective.
Sign in to post a comment and reply to other readers.
No comments yet
Be the first to share your perspective on this article.