This case study examines how AI tools respond to regulatory questions relevant to Stockbrokers and Trading Representatives practising in the United Kingdom. It draws on testing conducted against one regulation — the Financial Conduct Authority's Consumer Duty (PS22/9 and PRIN 2A) — spanning one aggregated question where AI assistants produced materially incorrect or misleading answers. The errors identified are not cosmetic: they concern the conditions under which a firm is or is not in breach of the Duty, a distinction that carries direct compliance and liability consequences.
Practitioners who rely on AI tools for regulatory guidance in this jurisdiction should treat these findings as a baseline signal about where AI-generated answers are most likely to lead them astray.
Stockbrokers and Trading Representatives in the United Kingdom routinely encounter Consumer Duty obligations when onboarding retail clients, designing or distributing investment products, producing suitability assessments, and handling complaints. When time is short — preparing a compliance note ahead of a client meeting, reviewing a terms document under deadline, or drafting a response to a regulator query — it is entirely natural to turn to an AI tool for a quick summary of what the Duty requires. Similarly, firms use AI assistants to train newer staff on regulatory basics or to generate first-draft policy language that a compliance officer then reviews.
The specific question tested here — what firms must do to prevent foreseeable harm and how customer-accepted risk alters that obligation — sits at the heart of Consumer Duty's outcome-based framework. For a stockbroker, this question arises every time a client is offered a higher-risk instrument and explicitly acknowledges the downside, or when a firm is deciding whether to proceed with an execution-only trade that it suspects may not suit the customer.
Getting the legal test wrong — whether a firm needs only a reasonable belief that the client understood and accepted the risk, or whether it must also satisfy a multi-condition checklist — can fundamentally alter how a firm structures its client documentation and conduct records.
If a practitioner acts on an incorrect AI answer here, the professional consequences can be severe. Stockbrokers operating under FCA authorisation face potential regulatory action — including variation or cancellation of their permission, public censure, or financial penalties — if their client processes are built on a misreading of the Duty. For the practitioner's clients, the stakes are equally real: a retail investor who was inadequately protected because the firm applied the wrong legal standard may suffer financial loss with limited recourse if the firm's own records appear compliant on the basis of a fabricated test.
Client trust, once broken by a compliance failure linked to avoidable error, is rarely rebuilt quickly.
Across the testing conducted for this case study, the pattern of error is consistent and specific: AI tools do not simply omit regulatory conditions — they substitute them. Where the FCA's Consumer Duty sets a single, deliberately bounded qualifier (whether the firm reasonably believes the customer understood and accepted a risk), AI assistants replaced that qualifier with a broader, multi-limbed test of their own construction.
The substituted conditions — good faith, supported understanding, avoidance of own-conduct harm, and general Duty compliance — are not without some grounding in the wider Consumer Duty framework, but they are not the legal standard for this specific rule. The effect is an answer that reads as authoritative and internally coherent while materially misrepresenting what the regulator actually requires.
The errors identified here cluster exclusively around the FCA's Consumer Duty, which is both the newest and the most behaviourally complex piece of retail regulation currently in force in the United Kingdom. This is not a coincidence. The Duty's outcome-based structure requires careful reading of individual rule provisions within a larger principled framework, and AI tools appear to conflate those two levels — applying the spirit of the broader Duty to a specific rule that has its own defined threshold.
This pattern of contextual over-generalisation is particularly hazardous for practitioners precisely because the resulting answer sounds more rigorous than the actual rule.
The systemic risk for Stockbrokers and Trading Representatives is that this type of error is effectively invisible without independent verification. A practitioner who asks an AI tool about customer-accepted risk under the Consumer Duty and receives a multi-condition answer has no obvious signal that anything is wrong — the response is detailed, cites the right regulation, and is broadly consistent with Consumer Duty principles.
The danger is that compliance processes and client documentation are then calibrated to a higher and non-binding standard, creating operational burden; or conversely that practitioners, believing they have understood the rule fully, do not verify it against the FCA handbook and miss the narrow, firm-centric test the rule actually applies.
1 finding in this case study. Click any to see its full evidence card.
Treat AI tools as a starting point for orientation, not a primary source for regulatory compliance positions. When a question involves the precise legal threshold that triggers or removes a firm's obligation — as is the case with customer-accepted risk under the Consumer Duty — the answer must be verified directly against the FCA Handbook or the relevant policy statement before it influences any client-facing process, internal policy document, or compliance record.
The failure mode identified in this case study is subtle: AI assistants produce answers that are thematically consistent with the regulation but apply the wrong standard at the rule level. That kind of error will not be caught by a quick plausibility check; it requires reading the actual provision.
Where AI use does influence a client deliverable or a compliance decision, maintain an audit trail. Note what question was asked, what the AI said, and what source was used to verify or correct the answer. This is not bureaucratic caution — it is the minimum needed to demonstrate to the FCA, in the event of a review, that your firm's processes were grounded in the regulator's actual requirements rather than an AI-generated summary.
Firms employing multiple stockbrokers or trading representatives should consider whether their current AI-use policy adequately addresses the risk of staff relying on unverified AI answers in regulatory contexts, and whether that policy has been tested against the specific areas where errors are most likely.
AI tools are most safely used in stockbroker and trading representative workflows for tasks that do not depend on regulatory precision: drafting non-regulatory client communications, generating first-draft questions for further research, summarising long documents that you will then verify against the source, or structuring the outline of a compliance memo before populating it with verified content. These are genuinely useful applications. The risk arises when practitioners skip the verification step because an AI answer looks confident and complete — which, in the case of Consumer Duty provisions, it often does even when it is wrong.
RegLeg's published Hallucination Research provides a free reference point for Stockbrokers and Trading Representatives who want to know — before acting on an AI answer — whether that area of regulation has already been identified as a high-risk zone for AI error. The research covers UK financial regulation including the FCA's Consumer Duty, and is structured so that practitioners can quickly locate findings relevant to their specific question or regulatory context. Checking the published research before finalising a compliance position informed by AI is a low-friction step that can prevent the kind of downstream error documented in this case study.
For firms employing multiple stockbrokers or trading representatives working across the same regulatory portfolio, RegLeg offers bespoke deep-dive reviews of specific regulations. These engagements map the precise points within a regulation where AI tools are most likely to produce plausible but incorrect answers, and translate those findings into practical guidance that can be embedded in a firm's existing compliance workflow. Where a firm's Consumer Duty programme, suitability framework, or execution-only policy relies on AI-assisted drafting, a targeted review can identify the highest-risk touchpoints before they become conduct issues.
RegLeg also produces training materials and CPD-aligned content designed specifically for practitioners who use AI tools in regulatory work. These materials focus on the failure modes that matter most in the UK financial services context — not generic AI literacy, but the specific patterns of error that recur when AI assistants are asked about FCA regulation. Separately, for firms that already have an AI-use policy in place, RegLeg offers a confidential review of that policy against its failure-mode catalogue, identifying gaps between what the policy assumes AI can reliably do and what the research shows it actually does.