Sponsored

UK financial regulation on AI moved from pilot language to delivery language in 2026.

On 21 April 2026, the FCA announced a second cohort for AI Live Testing. The list of firms and use cases read less like an experiment and more like a template of where AI is now embedded in real services: credit insights, AML detection, agentic payments, investment support, and customer-facing journeys.

At the same time, the FCA’s Mills review into advanced AI in retail services remains in focus, with feedback shaping recommendations due for summer delivery. Those two tracks — sandbox testing and structural review — now form a single compliance trajectory.

From advisory intent to operational evidence

In earlier phases, firms could point to AI policies and high-level risk statements. The Live Testing programme is changing that by requiring practical governance evidence under real workloads.

The regulator has positioned technical assurance and live monitoring as central. Its partnership with a specialist assurance provider indicates that policy language is now being paired with more explicit evidence standards: controls should be demonstrable in operating conditions, not just documented on paper.

For firms in financial services, this has three immediate implications:

  1. Use-case scope now equals accountability scope
    If AI is used in credit decisioning, customer support, transaction monitoring or onboarding, controls must be designed for those specific contexts.

  2. Supervisory expectations are becoming testable
    Firms are likely to face questions not about future roadmaps, but about incident logs, drift monitoring, override procedures and outcomes data.

  3. Existing principles may be tightened by practice
    The FCA has long relied on a principles-based framework, but principles still require interpretation. Live testing gives that framework a shared operating context.

Why this matters to consumers

The public impact is not a technical detail. AI decisions affect pricing, claims treatment, credit accessibility, and fraud response. If models are poorly governed, harms can spread rapidly across customer populations, and remediation often happens too late.

By putting AI systems into structured testing, the regulator gains insight into:

This can improve fairness and reliability, provided firms treat participation as a chance to improve internal controls rather than a box-ticking exercise.

The compliance lesson for UK finance executives

The combined read of the Live Testing updates and the Mills review is that AI compliance can no longer be an afterthought in governance committees. It now needs a standing seat with risk, legal, conduct and technology functions.

Executives should start with three minimum commitments:

Without this, firms may still pass static documentation checks but fail operational scrutiny when deployed systems are tested at scale.

What to expect next

The FCA has signalled a good-and-poor-practice report later in 2026. If that lands in published form, it is likely to reduce ambiguity across firms and speed up compliance benchmarks.

For now, the direction is visible:

UK regulators are not slowing innovation. They are setting a compliance floor for it.

Primary sources

Finance & Markets Correspondent
Covers: Finance, capital markets, technology investing

David Whitmore covers the intersection of capital and code — the funding rounds, market structures and policy moves that shape how money flows through the technology economy.