https://www.noxia.co.uk/field-notes/the-regulator-is-running-experiments · printed from noxia.co.uk · sources checked 23 September 2026
Field note · Buying AI
The FCA is testing AI with fifteen named firms, in public.
Regulators usually tell you what they think after the fact, in guidance written about a market that has already moved. This one is running supervised live tests with named firms on named use cases, and has committed to publishing what it found.
The short answer
The FCA’s AI Live Testing, part of its AI Lab, has run two cohorts. The first included Gain Credit, Homeprotect, NatWest, Monzo, Santander, Scottish Widows and Snorkl. The second, selected in 2026, includes Aereve, Coadjute, Barclays, Experian, GoCardless, Scottish Widows, UBS and Palindrome. Testing began April 2026, concludes at the end of 2026, and an evaluation report is due in Q1 2027.
On this page · 5 sections
The useful thing here is not the programme. It is the list of use cases, because it tells you what firms with compliance departments and real budgets have decided is worth testing under supervision.
The second cohort's cases span AI-enabled targeted support for investments, credit score insights for consumers, agentic payments, anti-money laundering detection and Know Your Customer. The technologies range from agentic AI and small language models to neurosymbolic approaches.
Three things a small firm can take from this without applying
Targeted support is on the list, and that is a signal. The regime went live on 6 April 2026 and firms are already testing AI-enabled versions of it under supervision. If you were treating targeted support as a distant possibility, the market is not.
The questions being asked are risk management and live monitoring, in the FCA's own framing — not whether the model is accurate. That is the regulator saying, in the design of its own programme, that the interesting problem is operating a system rather than choosing one.
An evaluation report is coming in Q1 2027. It will be the closest thing to a published account of how AI actually behaves in supervised UK financial services, and it will be free. Very little else on the market will be both.
What it does not mean
Being in a cohort is not approval, and nothing tested here becomes permitted by having been tested. It also skews large: Barclays, UBS, Experian and Lloyds are not a sample of the firms most people reading this run.
The first cohort had more of a spread — Monzo, Snorkl, Homeprotect, Gain Credit — which is worth knowing if you are wondering whether a smaller firm could ever be in one of these.
The part we would act on now
Whatever the report says in 2027, it will describe monitoring: what these firms watched, how they knew a system had drifted, what they did when it did. A firm that has none of that today will read the report as a description of somebody else's practice.
The cheapest version of it is not expensive. Record what the system decided and what a human changed. Watch the rate at which humans disagree with it. Keep what it refused to do. That is the whole of the audit layer, it is the argument in the trail note, and it is what Articles 22A to 22D ask for from a different direction. The targeted support regime itself is covered in the note on twenty-three million underserved consumers.
Questions people actually ask
What is FCA AI Live Testing?
Part of the FCA’s AI Lab, it is a supervised programme in which named firms test AI applications on live data with tailored support from the FCA and its partner Advai. Two cohorts have run; testing began in April 2026, concludes at the end of 2026, and an evaluation report is due in Q1 2027.
Which firms are in the FCA AI Live Testing cohorts?
The first cohort included Gain Credit, Homeprotect, NatWest, Monzo, Santander, Scottish Widows and Snorkl. The second includes Aereve, Coadjute, Barclays, Experian, GoCardless, Scottish Widows (Lloyds Banking Group), UBS and Palindrome.
What AI use cases are being tested?
AI-enabled targeted support for investments, credit score insights for consumers, agentic payments, anti-money laundering detection and Know Your Customer. The technologies range from agentic AI and small language models to neurosymbolic AI. All of them are customer-facing decisions with a regulatory consequence.
Does taking part mean the FCA approves an AI system?
No. Participation is supervised testing rather than authorisation, and nothing becomes permitted by having been tested. The cohorts also skew towards large institutions, though the first included several smaller firms.
Sources
- Financial Conduct Authority, “FCA announces second cohort for AI Live Testing”. Eight firms selected: Aereve, Coadjute, Barclays, Experian, Go-Cardless, Lloyds Banking Group (Scottish Widows), UBS and Palindrome. Applications opened January 2026; testing began April 2026; testing concludes at the end of 2026; evaluation report to be published in Q1 2027. Use cases span “agentic AI and small language models to emerging solutions such as neurosymbolic AI”, covering “AI-enabled targeted support for investments, credit score insights for consumers, agentic payments, anti-money laundering detection, and Know Your Customer”. Delivered with partner Advai. In the release’s own words, the initiative helps applicants “explore key questions around risk management and live monitoring to support the responsible deployment of AI for consumers and markets”; Jessica Rusu, quoted separately, refers to “tailored support from the FCA and Advai”. Published 21 April 2026. First cohort: Gain Credit, Homeprotect (Avantia Group), NatWest, Monzo, Santander, Scottish Widows (Lloyds Banking Group) and Snorkl. fca.org.uk ↗ — primary; the regulator’s own announcement of its own programme re-checked yearly
- The reading that the programme is about supervision rather than capability, and the three things to take from it without applying, are ours. — our own argument, labelled as such
Checked 23 September 2026. Next scheduled check 24 September 2027. Numbers that move — leaderboards, live indices — are re-checked every 30 days; annual datasets and rules in force every six months; dated research once a year. If something here has gone stale before we got to it, tell us and we will correct it and say what changed.
Cite this note
Noxia, “The FCA is testing AI with fifteen named firms, in public”, Field notes, 23 September 2026; sources checked 23 September 2026. https://www.noxia.co.uk/field-notes/the-regulator-is-running-experiments
Whatever the 2027 report says, it will describe monitoring.
Record what the system decided and what a human changed. Watch how often they disagree. Keep the refusals. We build that into whatever you already run, so that when the evaluation lands you are reading a description of your own practice rather than somebody else’s.
Talk to us about thisRead next
Advice & compliance
Targeted support went live in April. Twenty-three million people are the reason.
The FCA estimates 23 million consumers are underserved by the advice market. Targeted support is its answer, live since 6 April 2026, and it runs on segmentation a firm must be able to defend.
Advice & compliance
In a regulated firm, the audit trail is the product.
An unregulated business buys the output. You are buying the ability to explain the output eighteen months later to someone who is not impressed.
Advice & compliance
What your agent is allowed to decide on its own, since February.
Article 22 of the UK GDPR was replaced on 5 February 2026 by Articles 22A to 22D. The rule on automated decisions flipped from prohibition-with-exceptions to permission-with-safeguards.