The FCA published its second AI Live Testing cohort on April 22. Eight firms get to run AI systems against real customer data with the regulator looking over the shoulder: Barclays, Lloyds Banking Group (through Scottish Widows), UBS, Experian, GoCardless, Aereve, Coadjute, and Palindrome.
The use cases are not modest. Agentic payments. KYC. AML detection. Targeted investment “support” for retail customers. Consumer credit scoring. Each one a Consumer Duty landmine if it misfires, and the FCA is now sitting next to it while it runs live.
Applications opened in January, testing started this month, and the cohort wraps by year-end. The evaluation report lands in Q1 2027, with an interim “Good and Poor Practice” note before that. Advai, a London AI-assurance firm, is the technical partner.
Here’s the thing. The FCA still says it has no plans for bespoke AI rules. The Consumer Duty and SM&CR are the regime, full stop. Which means firms in the sandbox get something firms outside it don’t: a regulator who watched the model on live customers and didn’t object. That falls short of a rulebook, but it’s the closest substitute available before Sheldon Mills’ AI review reports to the FCA Board this summer.
For a Tier 1 bank running an agentic payments pilot, that is regulatory air cover the challenger banks can’t buy. For everyone else, the implicit question gets louder: did you run yours past the FCA, or not?
Cohort 1 launched October 2025. The Mills Review reports summer 2026.