This case study documents how AI tools respond to questions about Financial Conduct Authority regulation when consulted by Financial Advisers practising in the United Kingdom. One regulation was tested — the Consumer Duty (PS22/9 and PRIN 2A) — which sits at the centre of contemporary FCA supervisory expectations for retail financial services. Across ten aggregated question areas, AI tools produced incorrect, incomplete, or materially misleading answers in every case.
The errors span foundational legal questions about the Duty's origins, specific rule thresholds, and up-to-date supervisory developments — precisely the areas where a Financial Adviser is most likely to rely on AI for a quick authoritative steer. Taken together, these findings indicate that AI tools present a consistent and non-trivial compliance risk for Financial Advisers who use them to inform client advice, compliance documentation, or regulatory interpretation without independent verification.
Financial Advisers in the United Kingdom regularly encounter Consumer Duty questions in day-to-day practice: drafting suitability assessments and client-facing communications that must meet the consumer understanding outcome, reviewing fair value frameworks for products they distribute, assessing whether a particular client or entity qualifies as a retail customer, and responding to FCA data requests or supervisory queries. AI tools are increasingly consulted at exactly these moments — when a quick answer is needed and pulling up the FCA Handbook feels like the slower path.
The findings in this case study map directly onto those high-frequency use cases: the scope of the retail customer definition (including micro-enterprises and charities), whether consumer testing of communications is mandatory, how fair value assessments should treat non-monetary benefits, and which pre-Duty supervisory letters remain live expectations.
The professional stakes for a Financial Adviser who acts on an incorrect AI answer are significant. FCA Consumer Duty enforcement is outcomes-focused: if a firm cannot demonstrate that it assessed scope, value, and consumer understanding correctly, it faces supervisory action, skilled person reviews, and public censure regardless of how the error originated. Acting on a flawed AI answer about whether a client entity falls within the retail customer definition could mean delivering advice under the wrong regulatory framework — a regulatory breach with potential FCA investigation consequences, reputational damage, and exposure under the firm's professional indemnity policy.
Consumer Duty obligations also run through distribution chains, meaning that a Financial Adviser whose firm operates as a distributor carries responsibility for understanding the correct scope of their own obligations.
For the adviser's clients, the consequences of an AI-driven error are equally concrete. A client who is incorrectly told they fall outside the Consumer Duty's retail customer definition loses the protections — and the recourse — that the Duty affords. A client who receives advice built on a misunderstanding of the fair value framework, or on an outdated Dear CEO letter no longer in force, may be placed in unsuitable products or denied information they were entitled to expect.
Financial Advisers owe duties of care that extend beyond their own compliance: acting on an AI tool's inaccurate regulatory interpretation implicates both the client's outcome and the adviser's personal liability exposure.
All ten aggregated question areas produced incorrect AI responses, and the errors are not random — they follow identifiable patterns. The most common failure is condition substitution: AI tools reproduce the correct general principle but replace a specific qualifying term or threshold with a near-equivalent that changes the legal meaning. The clearest example is the charity threshold for the retail customer definition, where AI tools consistently substituted "annual income" for "annual turnover" — a distinction that matters in charity accounting and in the FCA's own drafting.
A related pattern is condition addition: AI tools append extra requirements to a rule that the FCA text does not impose, as occurred with the foreseeable harm question where the AI introduced multiple additional compliance prerequisites beyond the "reasonably believes" standard in the actual rule. Both patterns produce answers that sound authoritative and even more cautious than the rule requires, which makes them harder to identify as wrong.
A second cluster of errors involves more recent regulatory developments that sit at or beyond AI tools' knowledge boundaries. The FCA's withdrawal of more than 90 pre-Consumer Duty Dear CEO letters — announced in FS25/2 in March 2025 — was consistently mishandled: one AI tool fabricated incorrect retirement dates and split the event into two separate tranches, while another simply said the information was unavailable. Similarly, questions about first-year implementation speeches by senior FCA officials produced AI responses that blended distinct events into a single composite account, with source references that could not be verified against the actual speech record.
These are not obscure edge questions: they are the kind of supervisory currency that a Financial Adviser needs to be current on.
The systemic risk for Financial Advisers is that every one of these questions concerns a topic the adviser might plausibly consult an AI tool about, and in every case the AI's answer would lead the adviser to the wrong conclusion if taken at face value. The errors are not uniformly flagged as uncertain — several responses were expressed with the same confident register as a correct answer would be, including the citation of FCA Handbook URLs that appear authentic but which the AI used to support inaccurate claims.
A Financial Adviser operating under time pressure — which is the normal condition of practice — has no reliable in-response signal that the AI has gone wrong. Verification against the regulator's primary text is the only dependable safeguard.
10 findings in this case study. Click any to see its full evidence card.
Treat AI tools as a research starting point, not a primary regulatory source, for any question that touches Consumer Duty scope, rule content, or supervisory expectations. The findings in this case study demonstrate that AI tools can produce confident, well-structured, and superficially plausible answers on these topics while simultaneously misquoting thresholds, inverting the FCA's stated position, fabricating dates, and contradicting explicit rule exclusions.
No AI response on a Consumer Duty question should be incorporated into a client deliverable, compliance memo, or regulatory submission without independent verification against the FCA Handbook, the relevant policy statement (PS22/9), or the FCA's own published guidance (FG22/5). Where an AI tool cites a specific Handbook rule reference — such as a particular numbered provision — treat that citation as a claim to be checked, not a confirmed location.
Maintain an audit trail wherever AI tools contribute to regulatory work. If an AI-generated summary or rule description informed how you approached a question for a client, document what the AI said, when you checked it, and what the primary source confirmed. This discipline matters both for your own file management and for your ability to demonstrate to the FCA, in any supervisory review, that your advice had a verified regulatory basis. Never copy AI-generated regulatory citations — URLs or rule references — into a client document without visiting the cited source yourself.
The findings here include multiple instances where AI tools cited real-looking FCA Handbook URLs in support of an answer that contradicts what those pages actually say.
AI tools remain genuinely useful in Financial Adviser workflows for tasks that do not require regulatory precision. Drafting a first version of a non-regulatory section of client correspondence, generating a list of questions for further research, summarising a long document that you then read and verify, or producing a first-draft structure for an internal training slide deck — these are low-risk uses where AI can save time without creating compliance exposure. The line to maintain is between AI as a drafting and research aid on the one hand, and AI as a regulatory authority on the other.
For Consumer Duty questions specifically, the AI's track record on these tests makes that line worth enforcing consistently.
RegLeg publishes its hallucination research findings as a free reference for practitioners who want to understand where AI tools are most likely to go wrong on specific regulatory topics. For Financial Advisers working with the FCA Consumer Duty, the research documents the precise questions — and the precise failure modes — that AI tools have produced incorrect answers on.
Before relying on an AI answer about Consumer Duty scope, fair value methodology, the retail customer definition, or the current status of pre-Duty supervisory letters, advisers and their compliance teams can consult the published findings to understand whether that question has already been tested and what the verified position is.
For firms that employ teams of Financial Advisers working across the same regulatory portfolio, RegLeg offers bespoke regulation deep-dives tailored to the specific questions that team is most likely to put to AI tools. These sessions map the identified failure modes onto the firm's actual workflow, identify which AI use cases carry the highest verification burden, and provide a reference document the team can work from when AI tools are in use.
Firms building or reviewing their AI-use policy can also access a confidential review of that policy against RegLeg's failure-mode catalogue — a practical check that the policy addresses the specific patterns this research has identified, rather than relying on generic AI-use principles alone.
RegLeg also develops training materials and continuing professional development content designed for practitioners rather than technologists. The materials show Financial Advisers what a hallucinated regulatory answer looks like in practice, how to recognise the patterns — confident citation of incorrect thresholds, inverted regulatory positions, fabricated dates with plausible framing — and how to build a quick verification habit into normal workflow. If your firm is considering how to equip its advisers to use AI tools responsibly on regulatory questions, RegLeg is available to discuss what that looks like in the context of your specific practice.