Executive Summary
The CPMI-IOSCO 2026 consultation on initial margin transparency introduces specific public disclosure obligations for CCPs, including requirements around how margin model overrides are documented and communicated. For Company Secretaries advising CCPs or clearing members internationally, this consultation is live regulatory territory, responses were due in mid-2026 and implementation expectations are already hardening. Across the questions we put to AI tools on the disclosure obligations in this consultation, one aggregated finding emerged: a hallucination where AI produced a confident, plausible-sounding enumeration of disclosure requirements that the consultation text does not actually specify.
The failure mode is fabrication dressed as citation, the AI named specific categories of required disclosure (circumstances for override, authorised decision-makers, permissible adjustment types) where the consultation imposes only a broad general obligation. A Company Secretaries who took that answer at face value would have drafted advice premised on precision that the regulation does not provide.
How AI gets this regulation wrong
On this consultation, the dominant AI failure pattern is invention, the AI extrapolates granular, enumerated disclosure requirements from a single generic obligation, presenting the elaboration as though it were drawn directly from the regulatory text. The fabricated specifics are structurally plausible (the invented categories read like disclosure requirements that could exist), which is precisely what makes them dangerous. The table below maps where that failure lands.
| AI's Failure Mode | Count | Affected findings |
|---|---|---|
| Exposed Fabrication | 1 | Finding#1 |
What that means for your practice
For Company Secretaries, the risk from AI failures on this consultation concentrates in the regulatory enforcement dimension, specifically the exposure that arises when a CCP's board-approved disclosure framework is built around obligations that do not exist in the text. Because disclosure sufficiency is ultimately assessed by regulators against what the consultation actually requires, a Company Secretaries advising on framework design or sign-off needs the correct baseline, not AI's plausible reconstruction of it. The table below maps the risk impact of the finding in this cell.
| Risk Impact | Count | Affected findings |
|---|---|---|
| Regulatory enforcement | 1 | Finding#1 |
When this affects Company Secretaries
Company Secretaries touch this consultation most directly when advising a CCP's board on whether its existing public disclosure architecture satisfies the updated CPMI-IOSCO guidance, particularly on margin model overrides, where the board must be satisfied that what the CCP is documenting reflects both internal governance reality and regulatory expectation. A second common entry point is scoping the governance response to the consultation: mapping which disclosure gaps require a board decision, which require a policy amendment, and which are operational.
At both points, an AI-sourced summary of "what CCPs must disclose" functions as a baseline for materiality assessment, and a fabricated baseline produces a distorted materiality map.
The consultation imposes a general obligation: CCPs should publicly disclose relevant information on their override framework. That is the whole of it, at this consultation stage. A Company Secretaries who used an AI tool to characterise the specific required disclosure elements, and received the AI's invented three-item list, would be advising a board that precision obligations exist where only a principle-level obligation does. That affects not just the disclosure design but the board minutes, the audit committee sign-off language, and any regulatory correspondence asserting compliance.
The practical workflow risk is that AI's invented enumeration passes a basic plausibility filter. "Circumstances for override, authorised decision-makers, permissible adjustment types" sounds exactly like the kind of granular disclosure a consultation on override frameworks would require, which is why it does not immediately trigger a cross-check. A junior preparing a briefing note, or a Company Secretaries using the AI output as a first-pass read before the full consultation text is distributed, is the most exposed moment. The error travels through the document chain before it hits anyone with enough proximity to the primary text to catch it.
The findings at a glance
The table below summarises the one finding in this cell, the question asked, the AI's failure type, and the risk dimension it implicates for Company Secretaries advising on this consultation.
| # | Finding title | Type | Citation ID |
|---|---|---|---|
| 1 | CCP override framework disclosure: invented specificity | Hallucination | RLB-F-INT-BIS-CPMI-IOSCO-INITIAL-MARGIN-DISCLOSURE-CONSULT-2026-Q005 |
Aggregate impact
With one finding in this cell, there is no cross-finding pattern to aggregate, but the shape of the single error is instructive in its own right. The AI's failure is not a misread of the consultation text; it is an elaboration where no elaboration exists. The consultation deliberately leaves the override framework disclosure obligation at the principle level, "relevant information" is the operative phrase, and the consultation does not enumerate what that means. The AI filled that deliberate ambiguity with invented specificity, producing a three-item disclosure checklist that reads authoritatively and is entirely fabricated.
For Company Secretaries advising in multiple jurisdictions, that structural failure, AI converting principle-level obligations into enumerated specifics, is a recurring risk pattern on consultation documents, where regulators often use open-ended language intentionally. Consultations are not final rules. They invite responses partly because the obligations are not yet pinned down. An AI trained on finalized regulatory text may default to the kind of precision that final rules carry, misapplying that register to a document where the regulator's deliberate vagueness is itself the regulatory signal.
The systemic risk here is modest in absolute terms, one finding, one AI tool, one consultation. But the failure lands in a governance record-keeping context where precision matters disproportionately. Board minutes, sign-off language, audit committee papers, and regulatory correspondence all reference specific obligations. A Company Secretaries who has cited AI's invented three-item list in any of those documents has introduced an obligation that cannot be found in the source, and that is a different kind of exposure than a general misunderstanding.
What your team should do
The default position for Company Secretaries advising on this consultation is to go to the consultation text directly for any obligation-scoping work, particularly on override framework disclosure, where the gap between the principle in the document and the AI's invented enumeration is large enough to produce materially wrong advice. AI tools are a reasonable starting point for orientation on the consultation's structure, scope, and timeline, but the moment the work moves to specifying what a CCP must publish, AI's output needs to be treated as a first draft that requires primary-source verification.
For governance documents, board resolutions, audit committee papers, regulatory correspondence, that verification should happen before the document is circulated, not at review stage. The specific risk is that AI's invented disclosure categories are structurally plausible enough to survive a review by someone who has not read the consultation closely, which means the error can reach sign-off without triggering a challenge. The practical safeguard is to anchor every obligation statement to a specific passage in the consultation text, and to treat any enumerated list from an AI output as requiring a direct match to primary text before it is used.
Where AI tools remain useful for Company Secretaries work on this consultation: summarising the consultation's overall architecture and focus areas; identifying which sections are likely to require a board-level response versus operational implementation; and generating a first-pass gap analysis against existing disclosure frameworks, provided that gap analysis is validated against the consultation text before it is used in any governance document. The risk concentrates in the detail layer, and specifically in situations where AI converts a deliberately open-ended obligation into a false precision.
How RLB Can Help
RegLeg's published Hallucination Research gives Company Secretaries a practical pre-flight check before acting on AI-generated answers to regulatory questions. Each research entry documents the specific ways AI tools have misrepresented a regulation, wrong thresholds, fabricated obligations, outdated requirements presented as current, so that a Company Secretary can cross-reference those documented failure modes against any AI output before it reaches a board paper, a filing, or a governance record.
The research is freely accessible and structured around the failure types most relevant to secretarial practice: misstatement of procedural deadlines, incorrect attribution of disclosure obligations, and confusion between jurisdictional variants of the same rule.
For firms where multiple Company Secretaries work across a shared regulatory portfolio, RegLeg offers bespoke regulation deep-dives tailored to the specific instruments in scope. These engagements go beyond the published research to examine the precise provisions your team relies on most heavily, map the failure modes that carry the greatest secretarial risk for your firm, and produce a reference document your team can embed in its own AI-use workflow. The output is designed to be updated as regulations are amended, giving your team a living resource rather than a one-off snapshot.
RegLeg also develops training material and CPD-aligned content that equips Company Secretaries to recognise AI failure modes independently, not just to distrust AI output, but to interrogate it intelligently. Separately, RegLeg can conduct a confidential review of a firm's existing AI-use policy against its failure-mode catalogue, identifying where current controls adequately address known hallucination patterns and where gaps exist. Both services are delivered collaboratively, working alongside your governance and legal teams rather than as an external audit imposed on them.