This case study examines how AI tools perform when Compliance teams at Financial Advisory firms in the United Kingdom consult them on the FCA's Consumer Duty framework — specifically PS22/9 and PRIN 2A. Across nine aggregated questions put to AI assistants on this regulation, AI tools produced incorrect or materially incomplete answers in every case. The errors span core rule interpretation, threshold definitions, the distinction between binding rules and guidance, quantification requirements in value assessments, and the FCA's recent supervisory housekeeping activities.
Taken together, the findings indicate that Compliance teams relying on AI tools as a primary or authoritative source for Consumer Duty questions face a meaningful and systematic risk of acting on wrong information.
Compliance teams at Financial Advisory firms reach for AI tools in precisely the situations these findings concern: drafting internal policy and procedure notes on Consumer Duty obligations, building training and onboarding materials for advisers, conducting regulatory mapping when scoping new products or distribution arrangements, and responding rapidly to business-line queries about what the Duty requires in a specific situation. Where the firm is onboarding a new product provider or reviewing a distribution chain, Compliance may also use AI tools to generate initial due-diligence question sets against Consumer Duty criteria.
Each of these is a legitimate and time-efficient use of AI — but each of them is also a point at which a wrong AI answer becomes embedded in a firm work-product.
The corporate use-cases sitting on top of these topics are high-stakes. A policy note on Consumer Duty scope (who counts as a retail customer), drafted with AI assistance and circulated to the business, shapes who the firm treats as protected and how far it extends its harm-avoidance obligations. Fair value assessment methodology influenced by an AI tool's characterisation of the FCA's expectations will be reviewed by the FCA if it conducts a Consumer Duty thematic review or supervisory visit.
Compliance sign-off on communications based on a misunderstanding of whether consumer testing is mandatory or merely recommended sits directly in the line of FCA scrutiny. Training materials that misstate the legal basis of the Duty, or the carve-out where a customer understands and accepts a risk, may lead advisers to apply the wrong standard in client-facing situations.
For the firm, the cost of a wrong AI answer in these areas is not abstract. The FCA has wide powers under the Consumer Duty — it can require remediation at scale, impose financial penalties, require voluntary redress schemes, issue public censures, and in serious cases vary or cancel permissions. Beyond regulatory action, a Compliance team that has built processes on incorrect rule-interpretations may face significant internal remediation costs, disruption to business lines, and damage to the firm's relationships with the FCA and with professional indemnity insurers.
Individual employees are not typically held personally liable for acting on an AI tool's output, but the department, its leadership, and the firm carry the regulatory and commercial consequences.
Across all nine findings, the errors produced by AI tools fall into three recurring shapes. First, AI assistants silently drop critical qualifiers — the word "reasonably believes" becomes "the customer understands", a binding-rule distinction becomes a guidance recommendation, and the correct accounting measure (annual turnover) is replaced by a different one (annual income) without any signal that a substitution has occurred. Second, AI tools add conditions that are not in the regulatory text: they introduce extra requirements for firms to discharge before a carve-out applies, or they characterise the FCA's neutral stance on quantification as an active expectation.
Third, where the AI's training data contains gaps about specific FCA documents — in particular the FCA's FS25/2 statement of March 2025 and the detail of CP21/36-to-PS22/9 changes — AI tools either fabricate plausible-sounding facts (including specific dates and volumes) or state they cannot answer, without warning users that the gap exists. All nine findings concern a single regulation, the Consumer Duty (PS22/9 and PRIN 2A), which is the dominant regulatory priority for Compliance teams at Financial Advisory firms in the United Kingdom right now.
The systemic risk for a Compliance team is that these errors are not isolated events. A team using AI tools to support Consumer Duty compliance work will likely encounter multiple failure modes in the same week — a training deck that understates who counts as a retail customer, a policy note that imposes unwarranted quantification requirements in value assessments, and an internal briefing that mischaracterises consumer testing as a mandatory binding requirement rather than guidance. Each error is individually damaging; in combination, they represent a Consumer Duty compliance framework built on incorrect foundations.
The FCA's supervisory focus on Consumer Duty outcomes monitoring, fair value assessments, and communications effectiveness means each of these topics carries direct regulatory exposure.
Because AI tools typically present these answers with apparent confidence and cite FCA Handbook or guidance URLs, the outputs look authoritative to a non-specialist reader. That surface credibility is the core risk: a Compliance officer checking an AI-drafted policy against what appears to be an FCA Handbook citation has no reliable signal from the AI tool itself that the answer is wrong. The errors identified here require a reader already familiar with the correct rule text to catch — which is precisely the situation where AI tools are most likely to be consulted as a shortcut.
9 findings in this case study. Click any to see its full evidence card.
The starting position for any Compliance team at a Financial Advisory firm should be that AI tools are a research prompt, not a regulatory source. This applies with particular force to Consumer Duty questions: the errors identified across these findings are not edge cases or obscure technicalities. They affect core questions — who counts as a retail customer, what the carve-outs say, which obligations are binding rules and which are guidance, and what the FCA's current supervisory position actually is.
Until AI tools can reliably distinguish the precise text of FCA rules from plausible approximations of them, AI output in these areas should be treated as a starting-point for further verification, not as an answer that can be circulated or acted upon.
At the firm level, practical safeguards include a regulatory-verification policy that explicitly names Consumer Duty as a topic area where AI outputs must be checked against primary sources before being used in any firm work-product. Where AI output has influenced a policy note, training material, or internal briefing, that influence should be logged so the item can be re-checked if the underlying rule position changes. Sign-off requirements should apply before AI-assisted content enters firm-wide circulation: a Compliance professional with primary-source familiarity, not just familiarity with the AI's output, should confirm accuracy.
For regulatory-facing materials — fair value assessments, communications effectiveness records, scope analyses — AI-drafted or AI-summarised content should be clearly flagged internally so reviewers know to apply additional scrutiny.
AI tools are genuinely useful within the Compliance workflow for tasks that do not require the output to be accurate as a statement of regulatory law: drafting non-regulatory explanatory copy, summarising long FCA documents that the team will then read and verify, generating initial question lists for due-diligence or gap analysis, and structuring internal research briefs. These uses keep AI in a supporting role where its tendency to elide distinctions and add confident-sounding detail does less damage. The risk concentrates at the point where AI output is treated as the answer rather than as scaffolding for finding the answer.
RegLeg publishes its hallucination research findings as a free reference for Compliance teams who want to know, before they use an AI tool's answer, whether that question area has a track record of producing errors. For Consumer Duty questions specifically, this case study provides a direct pre-check: if the question your team is asking resembles one of the nine findings here, the research gives you specific information about how AI tools have failed on that question and what to verify against the primary source.
Using the published research as part of the team's AI-use protocol costs nothing and can prevent the kind of silent errors that are otherwise only discovered after a work-product has been circulated.
Where a firm wants a more structured picture of its exposure, RegLeg offers bespoke regulatory deep-dives tailored to the specific AI-supported workflows of Financial Advisory compliance functions. This work maps which areas of Consumer Duty practice — scope determinations, fair value methodology, communications compliance, supervisory correspondence — carry the highest hallucination exposure when AI tools are involved, and identifies which workflow steps pose the greatest risk of an AI error propagating into a firm-wide document or regulatory submission.
The output is practical: a prioritised list of the areas where the firm's existing AI use needs additional verification controls, and where current controls are sufficient.
For firms that have already developed an AI-use policy for the Compliance function, RegLeg can provide a confidential review of that policy against the failure-mode patterns documented in its research catalogue. This review identifies whether the policy's scope and verification requirements match the actual risk landscape, and produces a prioritised set of remediation steps where gaps exist. RegLeg also develops training and CPD-aligned content that Compliance teams can use internally — framed for advisers, paraplanners, and compliance support staff who use AI tools in their day-to-day work, not just for those with technical or regulatory specialist backgrounds.