AI mis-frames CPMI-IOSCO d226 effective practices as supervisory or mandatory obligations when the report itself describes them as non-binding examples of how PFMI standards can be met.
Executive Summary
Accountants supporting central counterparties, clearing members, and asset-management clients on variation margin control reviews are increasingly using AI to map d226 effective practices against client control libraries, draft compliance scoring matrices for internal audit, prepare regulator-readiness self-assessments for CCP governance committees, and validate disclosure footnotes that reference CPMI-IOSCO standards. Leading AI assistants tested by the RLB Specialist Panel produced confident, citable answers on the binding force of d226 that the document itself directly contradicts.
This cell collects one hallucination finding on the January 2025 CPMI-IOSCO publication "Streamlining variation margin in centrally cleared markets, examples of effective practices" (d226), organised for accountants working on CPMI-IOSCO VM matters in international and cross-border contexts. Across the finding, the AI subject in this audit produced a confident, citable response on the legal status of the eight effective practices in d226 that the document itself directly contradicts.
The Specialist Panel found that the AI classified each practice as a supervisory expectation or as carrying mandatory overlap with national rules, when the document's own stated purpose paragraph records it as "examples of how standards set out in the CPMI-IOSCO Principles for financial market infrastructures, as supplemented by the relevant guidance, can be met." Every finding in this cell is bound to verbatim regulator-issued source text held as primary substrate by the RLB Specialist Panel.
When this affects accountants work
This pattern surfaces whenever accountants use AI to characterise the binding force of d226 in control assessment matrices, internal audit work papers, and regulator-readiness self-assessments. The risk is structural: d226 is a CPMI-IOSCO publication that frames itself as voluntary effective-practice illustration, yet the AI's commitment reads as a supervisory or mandatory classification that a practitioner would paste into a deliverable before verification against the source. The deliverable then carries forward an inverted modality on binding force into downstream work products.
Findings overview
| # | Finding title | Type | Citation ID |
|---|---|---|---|
| 1 | Inverted modality on binding force | Misstated rule | RLB-H-INT-BIS-CPMI-CPMI-IOSCO-VARIATION-MARGIN-CCPs-2025-Q004-Opus47 |
What the AI got wrong
Finding 1: Inverted modality on binding force of d226 effective practices
Citation ID: RLB-H-INT-BIS-CPMI-CPMI-IOSCO-VARIATION-MARGIN-CCPs-2025-Q004-Opus47
In response to a Specialist Panel application-style question asking for a compliance obligations memo classifying each of the eight effective practices in d226 as either (A) MANDATORY REQUIREMENT, (B) SUPERVISORY EXPECTATION, or (C) VOLUNTARY GUIDANCE, the AI subject produced a complete memo that treated every one of the eight effective practices as either a supervisory expectation in its own right or as overlapping with mandatory national rules.
The memo opened with a threshold classification that placed d226 in category (C) but immediately added "a strong gravitational pull into (B) SUPERVISORY EXPECTATION because the underlying PFMIs are the de facto binding standard against which CCPs are supervised," then classified each practice individually under (B), (B) trending (C), (B) with (A) elements, or similar mixed labels.
The document's own stated purpose paragraph, by contrast, records that d226 sets out "examples of how standards set out in the CPMI-IOSCO Principles for financial market infrastructures, as supplemented by the relevant guidance, can be met." The text is illustrative of one way the underlying PFMI standards can be met, and is not a new layer of supervisory expectation or mandatory rule. The AI subject's framing inverts that modality: it converts a voluntary illustration into a baseline supervisory expectation, with each practice acquiring its own binding-force overlay.
The Specialist Panel records this as a misstated-rule finding because the AI's commitment on the legal status of d226 contradicts the document's own characterisation of itself. The full finding analysis, including the verbatim regulator text, the AI subject's complete answer, and the AI's failure mode classification, is at See full finding analysis.
Control assessment matrices and audit work papers that score clients against d226 practices as if they were supervisory pass/fail criteria distort residual-risk ratings, push remediation budgets toward voluntary illustrations, and risk under-rating exposures that arise from the underlying PFMI Principles or national rulebook overlays.
AI's failure mode
| # | Finding title | Type | Citation ID |
|---|---|---|---|
| 1 | Inverted modality on binding force | Misstated rule | RLB-H-INT-BIS-CPMI-CPMI-IOSCO-VARIATION-MARGIN-CCPs-2025-Q004-Opus47 |
Risk impact
| # | Finding title | Type | Citation ID |
|---|---|---|---|
| 1 | Inverted modality on binding force | Misstated rule | RLB-H-INT-BIS-CPMI-CPMI-IOSCO-VARIATION-MARGIN-CCPs-2025-Q004-Opus47 |
What accountants teams should do
- Anchor every d226-linked control reference to the document's own "examples of how standards … can be met" framing before mapping it into a control assessment matrix.
- Where a control reviewer or auditor scores adherence to a d226 practice, document explicitly whether the test relates to the practice itself (voluntary), the underlying PFMI Principle (binding under national supervision), or a national rulebook overlay.
- Avoid AI-generated phrasing that scores d226 practices as supervisory or mandatory expectations in work papers issued to client management or audit committees.
- When reporting d226-linked exposures up the chain, separate the binding-force question from the implementation-quality question so management decisions are based on the right legal characterisation.
- Build a short AI-output verification checklist for d226 work papers and require senior reviewer sign-off before any deliverable circulates outside the audit team.
Right of Reply
These findings and associated work have been put up in public with a view of the greater good for the development of a safer AI ecosystem. Any party reading this or any finding on reglegbrief.com may contact us and have an unconditional right of reply; the Specialist Panel will publish any factual correction or contextual response alongside the original finding, with no editorial gatekeeping. Researchers, regulators, and compliance teams with questions on methodology or specific findings can reach the Specialist Panel via the same channel.
Source & Methodology Standards
RegLeg Brief is operated by Verdus Technologies Pte. Ltd. (UEN 201616982R), incorporated in Singapore. The RLB Specialist Panel, with an aggregate of over 60 years of public-policy and industry experience, documents only confirmed hallucination findings, under a methodology that requires a verbatim regulator excerpt for every documented claim. All findings, citation IDs, model outputs, regulator excerpts, and methodology notes are open-access.
Primary source verified: d226 — Streamlining variation margin in centrally cleared markets — examples of effective practices (January 2025). R-folder reference: R6-FINAL_REPORT-00001. BIS portal: bis.org/cpmi.
Citation IDs referenced:
RLB-H-INT-BIS-CPMI-CPMI-IOSCO-VARIATION-MARGIN-CCPs-2025-Q004-Opus47