Lawyers advising on the Consumer Duty are increasingly using AI to draft client memos on Principle 12 scope, validate threshold language for fair value assessments, prepare partner-level briefings on PRIN 2A obligations, and reconcile FCA Feedback Statements such as FS25/2 against existing supervisory expectations. The work product is high-leverage: a single retail-customer-outcomes opinion can sit at the centre of a board paper, a Section 138D risk note, or an enforcement-defence brief.
Two frontier AI models tested by the RLB Specialist Panel produced 8 substantive failures on this regulation under audit conditions. The failure classes recorded are: Misstated Statutory Architecture, Inference Drift on the Foreseeable-Harm Safe Harbour, Confused Guidance with Rule on Consumer Testing, Hedge in Place of Verified FS25/2 Figure, Refusal to Confirm a Documented FS25/2 Count, Reversed the PRIN 2A Group-Insurance Exclusion, Invented Dual-Event Timeline for a Single FS25/2 Withdrawal, Refusal to Confirm FS25/2 Withdrawal Count.
Questions were prepared by the RLB Specialist Panel based on real practical AI usage in the workflows the respective audience uses AI for, and each finding is bound to verbatim regulator-issued source text held as primary substrate. The Consumer Duty (PS22/9 introducing Principle 12 and PRIN 2A, in force for open products from 31 July 2023 and for closed products from 31 July 2024) is the central retail-conduct regime the FCA now uses to grade firm behaviour, and the failure modes seen here all land inside the day-to-day work product that lawyers sign off on.
For lawyers, the operational consequence is direct. A Consumer Duty opinion or board memo built on the AI's framing imports a defect into client-facing work product. Cross-examination on the rule text, a Section 138D claim, or a supervisor's follow-up letter will surface the gap immediately, and the lawyer carries the professional liability.
Citation IDs for the findings in this brief: RLB-H-GB-FCA-CONSUMER-DUTY-PS22-9-Q002-Sonnet46, RLB-H-GB-FCA-CONSUMER-DUTY-PS22-9-Q003-Opus47, RLB-H-GB-FCA-CONSUMER-DUTY-PS22-9-Q007-Sonnet46, RLB-H-GB-FCA-CONSUMER-DUTY-PS22-9-Q013-Opus47, RLB-H-GB-FCA-CONSUMER-DUTY-PS22-9-Q013-Sonnet46, RLB-H-GB-FCA-CONSUMER-DUTY-PS22-9-Q018-Opus47, RLB-H-GB-FCA-CONSUMER-DUTY-PS22-9-Q020-Opus47, RLB-H-GB-FCA-CONSUMER-DUTY-PS22-9-Q020-Sonnet46. Each citation links to the per-finding record, the AI subject answer, and the regulator-issued substrate excerpt the answer was tested against. The RLB Specialist Panel maintains an audit-traceable record of which model produced which answer, against which substrate passage, and the binding is what makes the finding referenceable in firm work product and in supervisory correspondence.
The findings below are the ones that lawyers working under the Consumer Duty are most likely to encounter in the AI tools they already use, and the briefing sections that follow read each finding against the regulator-issued text.
This is the consolidated view of findings. Click the Citation IDs or 'see details →' on any item for the full details for each finding.
When advising a UK-authorised firm on the legal hierarchy that underpins the Consumer Duty, a lawyer who relies on the AI's framing risks documenting the wrong statutory ladder. FSMA 2000 is the source of the FCA's rule-making power, but the model's silent omission of FSMA 2023 leaves a partial picture that does not survive a regulatory mapping memo or a board-paper authority section. A Consumer Duty opinion that cites only FSMA 2000 misses the post-Brexit conduct architecture that sits alongside the Duty and exposes the firm to authority challenges in enforcement or supervisory correspondence.
Lawyers advising on the foreseeable-harm provision routinely draft retail-customer-journey opinions that turn on whether a single test applies (reasonable belief) or a composite multi-factor compliance check. The model substitutes the latter for the former, producing an over-broad standard that a litigator could not defend in a Section 138D claim and that compliance counsel could not align to PRIN 2A.2's actual text. Memos and risk-warning templates built on the AI's reconstruction expose the lawyer to professional liability when a client acts on what they believed was the rule.
A lawyer drafting an opinion on whether consumer-testing of communications is mandatory under the Consumer Duty cannot afford to conflate PRIN 2A.5's outcome-level obligation with FG22/5's guidance-level recommendation. The model's specific citation of PRIN 2A.5.10R as the binding test for consumer testing is a structural error: PRIN 2A.5 is outcome-prescriptive, FG22/5 is methodology-suggestive, and the two carry different enforcement consequences. Legal opinions that import the AI's mis-citation will mis-state the firm's obligations and create direct PI exposure.
Lawyers handling FCA supervisory or enforcement-defence work need accurate accounts of recent Dear CEO letter retirements; FS25/2 is the operative March 2025 publication. The model's fabricated April and August 2025 timeline is the kind of detail a supervisor's staff member will spot in three seconds, and a brief that recites the AI's account loses credibility immediately. The risk is not just being wrong; it is being demonstrably out of touch with the regulator's actual publication record.
Lawyers preparing FCA correspondence need to know which Dear CEO letters remain in force. The model's evasion response on a question the regulator's FS25/2 answers directly leaves the lawyer without a verifiable position, and the absence of a definitive answer in a regulatory submission is itself a problem. The lawyer who relies on the AI's hedged response will spend client time chasing a source that is on the FCA website in plain text.
Lawyers advising on Consumer Duty scope for insurance distribution, in particular the group insurance and large-risk commercial carve-outs in PRIN 2A.1.8R, are the population most exposed to the model's inverted answer. A scoping memo that brings group insurance distribution into the Duty's perimeter on the strength of a fabricated 'CP23/something' consultation paper is professionally indefensible and will be unwound by any specialist reader. PI exposure here is direct: the lawyer's advice imposes obligations on a client that the regulator has expressly excluded.
The same fabricated April and August 2025 timeline appears under a differently framed question, which is a strong signal for a lawyer that the model has an internalised but wrong account of FS25/2. A litigation or enforcement-defence brief that cites the AI's reconstruction risks being undermined by the regulator's own published record. The professional liability point is that this is not a one-off slip; the model reproduces the error consistently, so any brief, opinion, or supervisory letter that imports it will be wrong in the same way.
Lawyers reading the AI's Clifford Chance citation should treat the URL as fabricated until verified; the combination of evasion on a retrievable question and a manufactured supporting source is the most concerning pattern in the Sonnet 4.6 set. A legal memo that quotes the fabricated source as authority is a near-immediate professional-conduct issue if discovered by a reviewer, and the cost of independent verification on every Sonnet 4.6 citation undermines the productivity case for AI-assisted research.
Every finding on this page compares an AI subject's account of the rule against the regulator's verbatim text from the regulator's own portal. Both are linked. Each delta, its root causes, and impact analysis are documented and published with immutable Citation IDs.