Source-Credit Fabrication on CCP override framework disclosure. A frontier AI model, asked what CCPs must publicly disclose about their margin model override framework under the CPMI-IOSCO Initial Margin Disclosure Consultation, returned a confident three-part enumeration that does not appear in the consultative document and framed an expectation as a mandatory obligation. For a Compliance team at an investment bank, the misframing flows into CCP counterparty assessments, regulatory mapping, and supervisory submissions in ways that are difficult to unwind once the document is filed.

The pattern in one line

A frontier AI assistant, asked what the consultation requires CCPs to disclose about their margin model override framework, returns a closed three-element list, frames the obligation as mandatory, and supports the response with a secondary commentary URL rather than the primary BIS text, embedding a fabricated standard into the firm's CCP counterparty risk and supervisory work product.

How the RLB Specialist Panel tested this

Questions are prepared by the RLB Specialist Panel based on real practical AI usage in the workflows the respective audience uses AI for. The Panel binds each AI finding to verbatim regulator-issued source text held as primary substrate. For this finding, the Panel posed a Specialist Panel application-style question to two frontier AI subjects, framing the request the way a compliance analyst would type it into an AI assistant when scoping a CCP counterparty disclosure assessment for the prime brokerage or clearing operations desk.

The Panel then bound the model output against the verbatim consultation text on CCP override framework disclosure held as primary substrate, and against the broader d232 cover note on CCP margin model overrides.

The Panel does not give the AI subjects the substrate. The Panel evaluates only the AI subject's own output against the substrate the Panel holds. That asymmetry is what makes the finding usable: it shows how the model behaves when the analyst is not yet looking at the BIS d232 text.

What the models got wrong

Claude Sonnet 4.6, web search on, asked what specific information CCPs must publicly disclose about their margin model override framework under the 2026 consultation on initial margin transparency, returned a three-part answer: (1) the instances or circumstances where overrides may be warranted; (2) the key decision-makers authorised to exercise override discretion; and (3) the permissible types of adjustments that can be made.

None of those categories appears in the consultation. The consultation states: "CCPs should publicly disclose relevant information on their override framework." The specific content of "relevant information" is left open for comment. The model framed the disclosure as a "must", not a "should". It returned a closed three-part list, not an open expectation. It supported the response with a secondary commentary URL, not the primary BIS d232 cover note.

A compliance analyst who incorporates the three-part enumeration into the firm's CCP counterparty disclosure assessment template will measure every CCP counterparty against a standard the consultation does not set. A CCP that discloses general information on its override framework without enumerating the three fabricated categories will be flagged as deficient.

Citation: RLB-H-INT-BIS-CPMI-IOSCO-INITIAL-MARGIN-DISCLOSURE-CONSULT-2026-Q005-Sonnet46.

Why this matters for compliance officers

The downstream consequences for a Compliance function are concrete. CCP counterparty assessments record adequacy or deficiency against the assessment template. If the template is structured around the fabricated three-part standard, the assessment record overstates the obligation against which counterparties are measured. Remediation correspondence with CCPs, escalation notes to the credit risk and collateral management committee, and board-level updates on CCP counterparty risk all flow from the assessment record.

Where the Compliance team prepares a supervisory submission on the firm's CCP risk governance, the same misframing enters the firm's official record with its regulators. A submission that characterises the override framework disclosure as a binding "must" misrepresents the binding character of a consultative document. That kind of misstatement is the class of error supervisory teams look for when assessing the firm's understanding of its own obligations.

For an internationally active investment bank, the regulatory dialogue runs across multiple jurisdictions. A misframing in one supervisory submission propagates: home regulators, host regulators, and the firm's own group compliance reporting line all rely on the consistency of the firm's stated understanding. Unwinding the misstatement, once filed, is difficult; it requires a corrected submission and an explanation of why the original was incorrect.

The regulator's actual position

The consultative document states, verbatim: "CCPs should publicly disclose relevant information on their override framework." The cover note for d232 sets the override framework disclosure as one of several areas in which CCPs are expected to provide relevant information; it does not enumerate the contents of that disclosure, and it does not frame the obligation as a hard requirement.

The CPMI-IOSCO Secretariat's standard formulation on consultation-stage text is "should", not "must". The Secretariat reserves the prescriptive formulation for finalised standards. The decision to leave the specific content open for comment is itself a substantive feature of the consultation.

What this tells us about AI for compliance officers

The failure class is Source-Credit Fabrication. For Compliance teams, the operational signal is that any AI output reporting "the regulator requires (i), (ii), and (iii)" on a consultative document must be verified against the cited regulator text. The closed list structure conveys settledness; the verification pass must check whether the source actually contains a closed list at all.

The second signal is that the drift from "should" to "must" is the most common AI drift on consultation-stage CPMI-IOSCO and BCBS text. Compliance assessment templates that incorporate AI-drafted obligation language should be reviewed for the drift on every refresh.

The third signal is that secondary commentary URLs are not citation sources for compliance work product. A consultation released by the BIS Secretariat is the primary text. A third-party commentary is, at most, an interpretive aid. The verification discipline requires the analyst to follow the citation to the primary text on every assertion.

What the RLB Specialist Panel is doing about it

The Panel issues the finding bound to the verbatim consultation text held as primary substrate, with the citation ID assigned for cross-reference. The Panel makes the finding available to AI labs for incorporation into training-data and evaluation pipelines under the partnership track, and to enterprise Compliance functions as a worked example of the failure mode their AI-use controls must catch.

The Panel does not assess the model's overall accuracy on the consultation. The Panel assesses the specific output against the specific regulator text on the specific question a Compliance analyst is likely to ask in a CCP counterparty assessment workflow.

Citation: RLB-H-INT-BIS-CPMI-IOSCO-INITIAL-MARGIN-DISCLOSURE-CONSULT-2026-Q005-Sonnet46.

What compliance teams should do

Action items for a Compliance function at an internationally active investment bank using AI on CCP counterparty disclosure assessment and supervisory submission drafting:

• Verify every AI-returned enumeration of CCP disclosure obligations against the primary consultation text before the enumeration enters a CCP counterparty assessment template.
• Treat any AI output framing a consultation as imposing a "must" requirement as a presumptive drift: refer to the Secretariat text and confirm whether the formulation is "should" or "must" before any assessment record is finalised.
• Reject secondary commentary URLs as citation sources for supervisory submissions or board-level updates: if the AI cites a third-party commentary rather than the primary BIS text, treat the underlying factual claim as unverified.
• Build a pre-filing review step into the supervisory submission workflow that requires the drafter to identify, against the primary regulator text, the verbatim source for each obligation cited in the submission, and to record the verification in the submission's review log.
• Where a CCP counterparty assessment template is materially based on AI-drafted obligation language, refresh the template against the primary consultation text and document the refresh in the model risk and compliance assurance files.
• Use the RLB Specialist Panel's published findings as a working library of AI failure modes on regulator text, and brief Compliance analysts on the failure-class pattern as part of the onboarding for CCP counterparty risk work.

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: CPMI-IOSCO Consultative Report d232, Streamlining Variation Margin Disclosure (2026) · Substrate documents: p_03_ANNEX_Override_framework___public_disclosure_r_d232_covernote.pdf · CPMI portal: bis.org/cpmi

Citation IDs referenced:

• RLB-H-INT-BIS-CPMI-IOSCO-INITIAL-MARGIN-DISCLOSURE-CONSULT-2026-Q005-Sonnet46