Britain's Financial Conduct Authority (FCA) plans to automate much of its supervision and enforcement work using artificial intelligence, moving toward what it calls an "agentic regulator" — all without writing any AI-specific rules for the firms it oversees.

The approach was laid out at an evidence session before the House of Commons Treasury Committee on October 13 (full transcript).

How the FCA plans to become an "agentic regulator"

Jessica Rusu, the FCA's chief data, information and intelligence officer, told MPs that automation of the regulator's operations is expected to catch harms at the authorization gateway — the approval process firms must pass to carry out regulated activities in the UK. She said AI is already helping case officers read unstructured text and work through intelligence faster across authorization, supervision and enforcement, though a human is still signing off on every decision "for now."

Committee member John Glen asked whether the FCA would adopt the agentic supervision model recommended by the Mills Review. Rusu replied that the regulator was working toward becoming an "agentic regulator" — a model built on AI systems that can plan and carry out multi-step tasks on the regulator's behalf, rather than responding to one prompt at a time.

The Mills Review, published in July 2026 and led by former FCA executive director Sheldon Mills, recommends that the FCA build an AI-enabled agentic supervisory model and argues that the regulator needs greater powers. It also asked authorities to consider whether large language models (LLMs) such as ChatGPT, Claude and Gemini should abide by financial rules.

FCA chief executive Nikhil Rathi said one of the first jobs handed to the regulator's agentic tool was tidying regulatory returns, with the software going back to firms to flag numbers that do not add up. The aim, he said, is to free up human supervisors' time to spend on higher-risk cases.

Rathi, however, told MPs he could not agree that the FCA needs more powers until it has used its existing authority fully. He pointed to a change in the Financial Services and Markets Bill 2026 that shifts a Financial Ombudsman Service obligation from "may refer" to "must refer," which he said "could generate between 1,000 and 3,000 referrals."

Rusu also stated that frontier-model developers have been cooperating with the regulator, sharing details of model vulnerabilities and cyber risks that the FCA then passes on to financial firms.

No new AI rules planned

The FCA intends to rely on the AI frameworks already in place, especially the Consumer Duty, the set of standards requiring firms to deliver good outcomes for retail customers. It is an outcomes-based approach: firms are judged on the results they deliver to customers, not on the specific technology used to deliver them. In evidence to the Treasury Committee's January report on AI in financial services, Rusu said that with both the Senior Managers and Certification Regime and the Consumer Duty, the FCA does not "need to write new rules for AI."

The UK has no AI-specific legislation, and some 75% of UK financial services firms already use the technology.

The Mills Review highlights gaps in traditional provision: only 9% of consumers use traditional financial advice, around 900,000 UK adults are unbanked, and roughly £300 billion sits in low-interest accounts. One in five UK adults are open to letting AI make financial decisions for them, while about 26% already trust general-purpose chatbots for financial advice.

Rathi told MPs that it is actively being debated whether chatbots that give only general information, rather than advice for a specific person, should be regulated, adding that the decision is up to Parliament. That call would determine whether general-purpose chatbots sit inside or outside the FCA's regulatory perimeter.

Palantir contract ends; AI Live Testing findings due

Separately, the FCA's contract with Palantir, the US data-analytics firm, for financial crime work has ended.

Rusu is due to publish the FCA's response to frontier AI models in the week of October 19, alongside findings from its AI Live Testing program, which was announced in April. The second group to go through live trials with technical partner Advai included Barclays, Experian, Lloyds Banking Group's Scottish Widows and UBS. Testing concludes by year end, with an evaluation report due in the first quarter of 2027. Taken together, those dates set out when the FCA's next steps on AI will reach the public.