Financial regulation in the United Kingdom has taken a turn, as the Financial Conduct Authority is no longer confined to observing how regulated businesses deploy artificial intelligence and assessing the outcome afterwards. Through its AI Lab, the authority now supplies the computing power, the data environments and the models that firms use to build and test what comes next.
The Lab draws together five strands of work. The Supercharged Sandbox and AI Live Testing carry the practical weight, supported by AI Spotlight, the AI Sprint and the AI Input Zone, which gather ideas and evidence from across the market. The combination gives the regulator early visibility of new technology and gives firms a working partner at the point of construction.
Position matters a great deal here. Participants work through the hardest supervisory questions in a controlled setting with the regulator alongside them, and answers arrive in weeks instead of years. The conclusions reached during 2026 will inform how the rest of the market is assessed afterwards, which makes a place in the programme unusually valuable.
Five strands
The AI Lab is built as a set of connected routes into the same conversation. AI Spotlight surfaces the use cases already in development across the sector, the AI Sprint brings the industry together to work through shared problems, and the AI Input Zone collects views from firms, technology providers and consumer bodies on how supervision should function. Each feeds the two testing programmes that sit at the centre.
The design matters because it removes the usual sequencing problem. Evidence, market opinion and practical testing run at the same time instead of one waiting on another, so guidance arrives while the technology is still current. Firms with a view on how artificial intelligence should be treated have somewhere to put it, and the authority is building its position from live material.
Compute on tap
Cohort two of the Supercharged Sandbox launched in July and runs until 31 December 2026. It is built on NayaOne infrastructure with NVIDIA compute, and participants have been given access to Anthropic’s Claude, including Claude Code and Claude Cowork. A showcase demonstration day has been set for 26 November 2026, when the cohort presents what it has produced.
Claude Code and Claude Cowork extend the model into software development and multi-step task execution, which allows small teams to attempt work that would ordinarily require a much larger engineering function. Procurement cycles, infrastructure cost and access to frontier models have all slowed development across the sector, and the Lab absorbs each of them for the duration of the programme. Teams arrive with an idea and leave with something built, tested and understood by the authority that will supervise it.
Where the value sits
Five themes define the current cohort, and each maps onto a live commercial opportunity. Safer agent-led payments and commerce sits at the front, followed by fraud and economic crime detection, AI governance and accountability, access for underserved consumers, and compliance automation. The selection points at areas where better technology produces measurable return.
Agent-led commerce is the most consequential of the group. Software acting on a customer’s behalf to authorise and complete transactions opens a new category of financial product, and settling consent, authentication and liability inside a supervised environment gives participants a substantial head start on a market that barely exists yet. Fraud detection and compliance automation address costs every regulated firm already carries, so progress there improves margin as well as capability.
Governance and accountability sit slightly apart from the other four. Work on model oversight, audit trails and clear lines of responsibility is what allows the rest to reach production, since a board will approve an agentic payment flow only when it can explain how the system behaves and who answers for it. Access for underserved consumers points at demand the market has struggled to serve profitably until now, and lower marginal cost changes that arithmetic.
Names in testing
The second AI Live Testing cohort combines the largest institutions in the market with newer entrants. Aereve, Coadjute, Barclays, Experian, GoCardless, Lloyds Banking Group through Scottish Widows, UBS and Palindrome are all taking part. Testing concludes by the end of 2026, with an evaluation report due in the first quarter of 2027.
Retail banking, insurance through Scottish Widows, credit reference data, payments and property transaction infrastructure are all covered, giving the regulator a spread of use cases across very different risk profiles. Participants get something equally useful, which is sight of what works beyond their own sector and their own competitive set.
The evaluation report will carry real authority when it lands. It will be built on observed behaviour inside operating firms, across both incumbents and challengers, which puts it in a different category to guidance drawn from written submissions. Firms outside the cohort will be reading it closely in early 2027.
No new rulebook
The authority has confirmed it will not introduce AI-specific rules, relying instead on frameworks already in force. Outcomes-based regulation, on this reading, is capable of absorbing new technology without a dedicated rulebook, and firms remain accountable for results whatever tooling produced them. A report setting out good and poor practice is expected later in 2026, which will give the market a clear reference point without adding to the compliance burden.
Sheldon Mills is leading a review into the implications of advanced AI for consumers, retail markets and regulators, work that will define how far the existing perimeter reaches. The approach delivers a degree of certainty that has been scarce in other jurisdictions. Firms are not waiting on legislation before building, and capital can be committed with a much clearer view of the supervisory position.
Demand up 49%
Applications to the Regulatory Sandbox and Innovation Pathways have risen 49% year on year, a figure that reflects genuine confidence in the process. Firms are choosing to bring their most sensitive projects to the supervisor early, which suggests the route is understood as an accelerator and not an examination.
The regulator gains a broad and current picture of what the market is attempting, while participating firms gain proximity at the exact moment expectations are being formed. Both sides emerge better informed than any consultation cycle would allow, and useful guidance reaches the wider market sooner.
First mover clarity
Firms in live testing acquire something competitors cannot purchase, which is supervisory clarity ahead of deployment. Knowing how the authority assesses a governance framework, a model monitoring regime or an agentic payment flow before committing capital changes the economics of the whole project. It also shortens the distance between prototype and revenue.
Participants help shape the standards against which the wider market will later be measured, since the evaluation report and the good practice guidance will draw directly on what happens inside these programmes during 2026. Applying to the next cohort is a matter of positioning, and the opportunity is open to any firm prepared to bring a serious idea forward.