The FCA’s Mills Review: The AI Rulebook Isn’t Changing – But Your Firm Will Have To

On 6 July 2026, the FCA published the Mills Review, its long-anticipated examination of how artificial intelligence could reshape retail financial services in the UK by 2030. Led by Sheldon Mills and launched back in January, the Review drew on around 140 written submissions from firms, trade bodies, consumer groups and technologists, alongside dedicated consumer research. It is the most substantial statement yet of how the UK’s conduct regulator sees the AI-enabled future and, crucially, what it expects of firms operating within it.

The framing question the Review poses is a pointed one: not whether AI should be allowed into retail financial services, as Mills notes that ship has sailed, but rather who AI will serve as it becomes embedded in how consumers save, borrow, insure and invest.

The framework survives (and that’s the point)

If you were hoping for (or dreading) a new AI rulebook, the Review will disappoint you. Its central conclusion is that the existing regulatory framework remains sound. The FCA’s principles-based, outcomes-focused approach, anchored by the Consumer Duty and the Senior Managers & Certification Regime, was designed to flex across changing technologies and business models, and respondents to the Review overwhelmingly agreed. Notably, no firm argued that the senior manager accountability model should change to accommodate AI.

Regular readers will recognise the theme: we said earlier this year that no AI-specific rulebook is coming, and the Mills Review confirms it emphatically. But the corollary is just as important: if the rules aren’t changing, the burden of interpretation sits squarely with firms. What respondents asked for was not new regulation but clarity on how the existing regime applies to AI. As the FCA increasingly shies away from creating new standalone rules to add to an already complex and lengthy Handbook, it has preferred to publish clarifying guidance, often in the form of specific examples of ‘good’ and ‘poor’ practice; expect this to follow later this year or early next year.  Until that guidance arrives, demonstrating that your governance maps AI risk onto Consumer Duty outcomes and SM&CR accountabilities is your problem to solve, not the FCA’s.

The autonomy spectrum: a governance tool you can borrow

The most practically useful contribution of the Review is its five-level “autonomy spectrum”, describing the shifting role of the human as AI takes on more of the work: operator, collaborator, consultant, approver and, ultimately, observer.

This matters because the risks change character as you move along the spectrum. At the operator and collaborator end, the questions are about accuracy, over-reliance and the quality of AI-assisted outputs. At the approver and observer end, the questions become harder: consent, delegated authority, accountability when an autonomous agent transacts on a consumer’s behalf, and redress when something goes wrong.

Our suggestion: borrow the spectrum, and use it as your taxonomy for AI risk. Mapping every AI use case in your firm against these five levels is an excellent starting point for a proportionate, risk-sensitive governance framework, and a helpful way to build out your internal AI risk register. It’s also a far more defensible one than treating “AI” as a single undifferentiated risk category.

What it means for insurance

For insurers and insurtechs, the Review has plenty to say. It identifies insurance as a market where AI could accelerate embedded distribution, automated quote comparison, claims triage and platform-led steering of product discovery. It is equally candid about the risks: hyper-personalisation promises better matching of products to customer needs, but brings with it the potential for bias, opaque pricing and personalised manipulation. The Review specifically flags the risk of AI models using proxies for protected characteristics in pricing decisions; a concern that should already be on the radar of every pricing and underwriting function using machine learning.

For firms in this sector, the message is that Consumer Duty expectations around fair value, customer understanding and support do not dilute as decisions become more automated. If anything, the evidential bar rises: you will need to show how you know your AI-driven pricing, distribution and claims decisions are delivering good outcomes.

Governance is the competitive edge

Perhaps the most striking strategic point in the Review is its framing of governance as an enabler rather than a brake. Firms that can demonstrate auditability, robust testing, clear permissions, live monitoring and effective escalation will be able to deploy AI more confidently, more quickly and at greater scale than those that cannot. In an AI-enabled market, governance maturity becomes a source of competitive advantage.

The Review also signals that model risk management must extend beyond point-of-deployment validation into continuous monitoring for drift and degradation, and that dependency on third-party model providers (often a small number of very large ones) is now a core governance question, not something for procurement to note and forget about.

What happens next

The Review closes with seven priority recommendations to the FCA Board, spanning the regulatory perimeter, system-wide oversight of AI concentration risks, monitoring the market’s transition along the autonomy spectrum, scaling the AI Lab and sandbox work, laying the foundations for “agentic finance”, building an Agentic Supervisory Model for the FCA itself, and exploring a public-interest financial capability service. None of these is a rule change today, with any decisions on adoption now resting with the FCA Board, but the direction of supervisory travel is unmistakable, and the Treasury Committee is still expecting practical guidance from the FCA by the end of the year.

Don’t wait for the Board

Firms do not need to wait for the FCA’s formal response to act. A gap analysis against the autonomy spectrum, an AI use-case register with clear senior manager ownership, updated Statements of Responsibilities, and a monitoring approach that treats AI models as living systems rather than one-off deployments are all no-regrets moves you can make now.

If you would like help developing a proportionate, risk-sensitive AI governance framework, or simply want to talk through what the Mills Review means for your firm, please get in touch.