UK's FCA asks AI chatbots to carry money-advice warnings as it turns itself into an "agentic regulator"
What's going on? Britain's Financial Conduct AuthorityDealroom has a profile for this one. Try Dealroom → plans to become an "agentic regulator", using AI to automate much of its supervisory and enforcement work, the Financial Times reported on 15 October. Chief data, information and intelligence officer Jessica Rusu told MPs on the Treasury Committee on 13 October that the watchdog wants "to automate wherever we can possible and to intervene at scale", and to "shift left" harms it can spot at the authorisation gateway.
Update, 19 October: Setting out the FCA's response to the Mills Review at Bloomberg's UK Policy Series, Rusu said the regulator is embedding six of the review's seven recommendations into live work immediately, with the focus on testing, assurance and monitoring alongside firms rather than further consultation. She called on AI providers to give financial information and advice generated by their models "clear, proportionate and consistent disclaimers", noting that consumers increasingly turn to ChatGPT, Gemini and Claude for money questions. Four in five less-experienced investors aged 18 to 40 have used AI to help with investing, but only 40% of consumers know there is no recourse if advice from a general-purpose AI tool goes wrong. Rather than build the new public financial-capability tool the review suggested, the FCA will build on existing work, including IOSCO's first TechSprint.
Good and poor practice: The FCA also published its first good and poor practice examples for AI in financial services, drawn from AI Live Testing, the Supercharged Sandbox and supervisory work. Good practice included a firm whose chatbot uses a live list of vulnerability signals to pass customers to human advisers, and an independent ethics panel that vets AI use cases before deployment. Poor practice included building AI tools without a defined consumer need, overstating capabilities, retrofitting governance after deployment and assuming a large language model can reliably spot vulnerable customers without proper testing. The examples create no new rules: firms remain responsible for AI outcomes under the Consumer Duty, and the FCA says it will update them as the technology evolves.
The regulator's own agents: Rusu said the FCA is building AI tools to oversee the 60,000 anti-money-laundering firms coming into its remit in 2028, and a "market first responder" agent to supervise around 9,000 small wholesale transaction-reporting firms. After launching a machine-readable Handbook API in August, it is working on machine-readable data-collection APIs, and it will work with industry and frontier labs on a framework for agentic finance covering agent identity, delegated authority and accountability. She also said incident reporting must extend to significant AI failures and near-misses. According to the speech, a joint FCA and Bank of England survey due next week shows 86% of firms now use AI.
What the FCA is already doing: Rusu said AI handles unstructured text well and is already helping case officers analyse intelligence faster across authorisation, supervision and enforcement, with a human still responsible for every decision "for now". Chief executive Nikhil Rathi said one of the first uses of its agentic tool was cleaning regulatory returns, going back "agentically" to firms to flag submissions that do not reconcile, freeing supervisors for higher-risk work. Rusu said frontier-model developers have been cooperating with the regulator, which shares information on model vulnerabilities and cyber risk with financial firms.
Context: The push follows the Mills Review led by former FCA executive Sheldon Mills in July, which recommended an AI-enabled, agentic supervisory model, argued the watchdog needs greater powers and urged authorities to look at whether ChatGPT, Claude, Gemini and other large language models should fall under financial rules. According to the FT, research for the review found a fifth of UK adults would be open to letting AI models make financial decisions for them. Rusu set out the FCA's formal response on 19 October (see the update above).
No new AI rulebook (yet): The FCA is stopping short of AI-specific rules and is leaning on existing frameworks such as Consumer Duty. Rathi told MPs he is not convinced the FCA needs "lots more powers" until it has shown it has used its current ones to the fullest. Chatbots that give general information rather than individualised recommendations currently sit outside the regulated perimeter, a boundary he called "a very live discussion"; whether LLMs themselves should be regulated is ultimately a matter for Parliament.
Why it matters: One of the world's most closely watched financial regulators is turning AI on its own supervision rather than writing new AI rules for the industry. For UK fintechs and incumbents, that points to faster, data-driven scrutiny of filings and conduct, while the open question of whether consumer AI assistants giving money advice need regulating hangs over wealth and advice start-ups. The 19 October package confirms the approach: published examples and a call for chatbot disclaimers rather than new rules, plus AI agents supervising thousands of smaller firms.
Sources: FCA speech by Jessica Rusu (Bloomberg UK Policy Series, 19 Oct 2026); FCA, AI in financial services: good practice and areas for improvement (19 Oct 2026); FTAdviser (19 Oct 2026); Financial Times (Martin Arnold, Emma Dunkley, 15 Oct 2026); Compliance Corylated (Treasury Committee session, 13 Oct 2026); Treasury Committee transcript via Parallel Parliament (13 Oct 2026)
