This case study examines how AI tools perform when Accountants (CA/PA) practising in the United Kingdom ask questions about Financial Conduct Authority regulation — specifically the Consumer Duty framework introduced under PS22/9 and PRIN 2A. Across two aggregated question areas, AI tools produced materially incorrect or misleading responses in every instance tested. The errors were not minor ambiguities: AI tools substituted incorrect legal thresholds, omitted mandatory qualifiers, and in one case directly contradicted the FCA's published position on what firms are required to do.
For accountants advising clients on Consumer Duty compliance, or reviewing whether a firm falls within the Duty's scope, these failures carry direct professional and regulatory risk.
Accountants (CA/PA) in the United Kingdom regularly encounter Consumer Duty questions in the normal course of practice. This may arise when advising a financial services client on whether the Duty applies to a particular product line or customer segment, when reviewing a client's fair value assessment framework ahead of an FCA review, when drafting compliance opinions or board-level memos on regulatory scope, or when onboarding a new financial services engagement and needing to establish which obligations attach.
It also arises in staff training contexts, where an accountant may use an AI tool to quickly surface a summary of the rules before preparing training materials.
In all of these situations, the accountant is acting as a trusted expert. Clients and colleagues rely on the accuracy of regulatory guidance provided — and in many cases that guidance will flow directly into compliance documentation, board papers, or client-facing advice without being independently cross-checked against the primary source. The AI tool's response therefore does not remain with the accountant; it propagates into deliverables.
If the AI's answer is wrong, the consequences can be severe on both sides of the relationship. For the practitioner, reliance on an incorrect AI summary of FCA rules without independent verification may constitute a failure of professional competence, exposing the accountant to complaints, professional discipline, or malpractice claims. For the client, acting on incorrect guidance about Consumer Duty scope or fair value methodology could result in a non-compliant programme, FCA supervisory action, financial remediation obligations, or reputational harm — all of which can rebound on the accountant who provided the underlying advice.
Across both question areas tested, AI tools produced errors that fall into a consistent pattern: they replaced precise regulatory language with approximate or looser equivalents, and they inverted or softened mandatory obligations into conditional ones. In the first finding, AI tools substituted the defined threshold term used in the FCA Handbook with a different legal concept, and framed a general Handbook definition as applying only to specific activity types. In the second, AI tools recast a clear statement that the FCA does not expect quantification into an encouragement of quantification where practicable — directly reversing the regulator's published position.
Neither error was trivial; both touched the specific legal terms and obligations that determine whether a firm is compliant.
Every finding in this set relates to the same regulatory instrument: the Consumer Duty as set out in PS22/9 and PRIN 2A. This is not a scattered pattern across multiple regimes — it is a concentrated set of failures within a single, high-profile and actively enforced FCA framework. The Duty has been a primary supervisory focus for the FCA since its implementation in July 2023, and firms are regularly asked to evidence their compliance. Errors on scope (who the Duty applies to) and on methodology (what the fair value assessment must include) go to the heart of that compliance picture.
The systemic risk for an Accountants (CA/PA) relying on AI tools in this area is therefore high. In both question areas examined, the AI produced responses that appeared authoritative — citing FCA Handbook provisions and finalised guidance — while the substance was wrong. An accountant who did not independently verify the AI's answer against the primary regulatory text would carry an incorrect understanding into their work. Given that these questions arise in scope assessments, compliance reviews, and fair value sign-off processes, the probability that an unchecked AI error reaches a client deliverable is significant.
2 findings in this case study. Click any to see its full evidence card.
The default position for any Accountants (CA/PA) using AI tools on Consumer Duty questions — or on FCA regulatory topics more broadly — should be that the AI output is a starting point for research, not a primary source. The errors identified in this case study occurred on questions that appear factual and bounded: who does the Duty cover, what does a fair value assessment require. These are exactly the kinds of question where AI tools can produce a confident, well-formatted answer that is nevertheless wrong in a legally material way.
Treat AI-generated regulatory summaries the way you would treat a colleague's verbal recollection: useful for orientation, not sufficient for sign-off.
For any work that will influence a client deliverable, board paper, compliance opinion, or staff training material, two practical safeguards should be standard. First, verify the AI's answer directly against the regulator's published text — the FCA Handbook and finalised guidance are publicly accessible, and the specific provisions are short enough to read in full. Second, maintain an audit trail of any AI use that contributed to a regulatory conclusion: note what was asked, what the AI said, and what primary source confirmed or corrected it.
If a complaint or professional review arises later, that trail demonstrates the accountant applied independent judgement rather than delegating to an AI.
There are areas of Accountants (CA/PA) work where AI tools add genuine value without these risks: drafting routine non-regulatory correspondence, generating initial checklists or question lists for a new engagement, summarising lengthy documents that you will then verify independently, or producing first-draft narrative sections of reports where the regulatory content will be separately reviewed. The risk concentrates when AI output goes directly into regulatory conclusions without a human review step against primary sources.
RegLeg's published Hallucination Research is available as a free reference for Accountants (CA/PA) who want to check a regulatory topic before acting on an AI answer. The research covers specific questions across the FCA's regulatory handbook — including Consumer Duty provisions — and identifies where AI tools have produced materially incorrect responses. Before relying on an AI summary of a Consumer Duty obligation, an accountant can check whether that question area has already been examined and whether a known failure mode applies.
For firms that employ multiple accountants working across the same regulatory portfolio — for example a practice with a financial services team advising on Consumer Duty compliance across a number of clients — RegLeg offers bespoke regulation deep-dives. These are structured reviews of a specific regulatory instrument, mapping the question areas where AI tools are most likely to mislead and producing a reference document the team can use as a check against AI output. This is particularly relevant for Consumer Duty, where the FCA's supervisory focus means errors have a high probability of being tested.
RegLeg also works with firms on training materials and CPD-aligned content that show Accountants (CA/PA) the specific failure modes to watch for when using AI tools on regulatory questions: threshold substitution, obligation inversion, scope narrowing. For firms that have already developed an internal AI-use policy, RegLeg can review that policy against its failure-mode catalogue on a confidential basis, identifying gaps before they become professional or regulatory exposure. The aim in all of this is a practical working relationship — not a replacement for the accountant's own judgement, but a structured resource that makes that judgement better informed.