The FCA has been unambiguous about one thing throughout 2026: it is not planning to write a bespoke AI rulebook. Instead, its approach — set out by Alex Smith, the FCA’s Head of Cross-Cutting Policy and Strategy, in the FCA’s 8 June 2026 blog — rests on applying existing frameworks, including the Consumer Duty, SM&CR and existing governance and control expectations, to AI use cases as they emerge, while running a structured engagement programme through the AI Lab to learn how firms are actually deploying the technology. 

WHAT’S HAPPENING NOW 

In April 2026, at UK FinTech Week, the FCA confirmed the second cohort of firms admitted to its AI Live Testing service — a mix of established names including Barclays, Experian, UBS and Lloyds Banking Group’s Scottish Widows alongside newer entrants in compliance and property technology. Applications this round reflected how quickly the underlying technology is moving: firms are testing agentic AI, small language models and emerging neurosymbolic approaches, across use cases spanning AI-enabled targeted support for investments, credit score insights, agentic payments, anti-money laundering detection and Know Your Customer processes. 

Testing began in April and runs to the end of 2026, with an evaluation report due in Q1 2027. The FCA has also kept an “AI Input Zone” open — a standing survey inviting firms to flag where regulatory clarity would actually help — which closed for this round on 19 June 2026 and will feed into a good-and-poor-practice publication expected later this year. 

WHY “NO NEW RULES” DOESN’T MEAN “NO NEW EXPECTATIONS” 

It would be a mistake to read the FCA’s reluctance to legislate as a lack of interest in how firms are governing AI. Our reading of the FCA’s approach is that firms using AI will still need to apply the relevant existing standards to those use cases — which in practice means being able to explain, evidence and monitor what their models are doing, using the same governance discipline they would apply to other regulated decision-making. 

For AML and KYC use cases specifically — two of the areas explicitly named in this cohort’s testing — that means demonstrable bias testing, clear escalation paths when a model flags something ambiguous, and an audit trail that a supervisor could pick apart after the fact. “The model decided” is not going to be an acceptable answer to a Nationwide-style enforcement question in three years’ time. 

WHAT THIS MEANS IF YOU’RE NOT IN THE COHORT 

Most regulated firms won’t be part of AI Live Testing, but the direction of travel still applies to them. The parallel Mills Review, launched in January 2026 and published in July, examined how advanced AI could reshape retail financial services, including the implications for firms, consumers, markets and regulators. Alongside the forthcoming good-and-poor-practice publication, it gives firms another important indication of where the FCA sees AI-related risks and opportunities developing. 

GOVERNING AI DEPLOYMENTS WITH REGTECH 

Firms adopting AI in risk, compliance or customer-facing functions need the same discipline the FCA is asking live-testing participants to demonstrate: clear ownership, continuous monitoring, and evidence that stands up to scrutiny. The Axiol platform supports this by: 

• Treating AI-driven processes as risks in their own right, scored and tracked alongside every other control in the enterprise risk framework 

• Assigning named ownership to AI-related risks, so accountability doesn’t disappear into “the algorithm” 

• Capturing an ongoing audit trail of model governance activity, ready to support supervisory questions or good-practice benchmarking 

• Linking AI risk evaluations to the wider control framework, so weaknesses in one area — bias, explainability, data drift — are visible wherever they have downstream impact 

The FCA’s AI Lab is ultimately a preview of what “good AI governance” will be measured against across the whole market. Get in touch to find out how Axiol can help your firm build that evidence base now.