Failure classes on record for Compliance teams at Corporate Banking firms: Exposed Fabrication.

Frontier AI models tested by the RLB Specialist Panel on the December 2025 CFTC final rule on swap dealer business conduct and documentation produced one discrete hallucination bound to verbatim regulator-issued source text. Each finding has a direct read-through into compliance teams at corporate banking firms' working deliverables on swap dealer business conduct compliance.

The pattern in one line

Leading AI assistants used by compliance teams at corporate banking firms on the December 2025 CFTC final rule on swap dealer business conduct and documentation returned answers that looked sourced and coherent but conflicted, in load-bearing specifics, with the regulator's verbatim text on appendix identity, staff no-action letter scope, or the product boundary of the pre-trade mid-market mark requirement.

The errors survive a first-pass review of compliance teams at corporate banking firms' written supervisory procedure update, compliance attestation, audit walkthrough documentation, or supervisory communication on swap dealer business conduct and only surface when a counterparty, regulator, or independent reviewer checks the primary record.

How the RLB Specialist Panel tested this

The Panel prepares questions based on real practical AI usage in the workflows corporate-banking compliance officers use AI for: drafting, validation, benchmarking, and source-citation tasks against the December 2025 CFTC final rule on swap dealer business conduct and documentation. Each tested question is bound to verbatim regulator-issued source text held as primary substrate; the Panel does not generate findings against documents whose verbatim text it cannot anchor. For this cell, the Panel ran one Specialist Panel question against two frontier AI subjects with web search on, and recorded the AI's confident answer alongside the regulator's actual text for direct comparison.

Where a subject retracted only under direct challenge, the Panel records that as Exposed Fabrication; where the subject confirmed an event without giving the load-bearing detail the question asked for, the Panel records that as Inference Drift.

What the models got wrong

Finding 1 (RLB-H-US-CFTC-SWAP-DEALER-BUSINESS-CONDUCT-DOCUMENTATION-2025-Q004-Opus47, Claude Opus 4.7 (web search on)). On the Specialist Panel question covering which categories of swap instruments were actually subject to the pre-trade mid-market mark disclosure requirement under § 23.431(a)(3) prior to its elimination, and whether 'eliminated in its entirety' means the requirement was removed for all swap types, the AI returned: "PTMMM disclosure is no longer required for any swap subject to § 23.431, including cleared credit default swaps (index and single-name), with the exemption product-agnostic across the desk's covered swap book." The regulator-issued source text, held by the Panel as primary substrate from the CFTC final rule as published at 90 FR 61226, records: "the Commission eliminates the PTMMM Requirement in its entirety by deleting paragraphs (i) and (ii) of § 23.431(a)(3), but § 23.431(c) already excluded cleared swaps and execution-venue-initiated transactions from PTMMM's reach, so cleared CDS were outside scope before the amendment." The Panel classifies this as Exposed Fabrication.

Why this matters for compliance teams at corporate banking firms

Compliance teams at Corporate Banking firms working on the December 2025 CFTC final rule on swap dealer business conduct and documentation carry a direct read-through from the AI's wrong answer to the written supervisory procedure update, compliance attestation, audit walkthrough documentation, or supervisory communication on swap dealer business conduct they sign off on. The exposure is examination findings, remediation across training and counterparty communication templates, and a paper trail of internal documentation that misrepresents the regulatory baseline.

The failure modes recorded here touch the load-bearing compliance and documentation specifics that compliance teams at corporate banking firms are paid to get right: which appendix the January 2026 correction restored, which trading venues a CFTC staff letter actually covers, and which swap product types were within the scope of § 23.431(a)(3) before its elimination. Where an AI assistant returns a confident, plausible-looking answer that conflicts with the regulator's verbatim text on any of these dimensions, the cost of correction rises with every downstream artefact that cites it, and the firm's regulatory exposure compounds across the swap dealer documentation stack.

The regulator's actual position

On which categories of swap instruments were actually subject to the pre-trade mid-market mark disclosure requirement under § 23.431(a)(3) prior to its elimination, and whether 'eliminated in its entirety' means the requirement was removed for all swap types, the CFTC final rule as published at 90 FR 61226 records:

Paragraph (a) of § 23.431 shall not apply with respect to a transaction that is initiated on a designated contract market or initiated with a counterparty whose identity is not known to the swap entity prior to execution on a swap execution facility. The December 2025 final rule deletes paragraphs (i) and (ii) of § 23.431(a)(3) and moves the price disclosure and compensation disclosure requirements into paragraphs (2) and (3) of § 23.431(a).

Each verbatim block above is held by the Panel as primary substrate and is the anchor against which the AI subjects' answers were compared. The Panel does not generate findings against documents whose verbatim text it cannot anchor.

What this tells us about AI for compliance teams at corporate banking firms

The pattern recorded against this cell maps to the failure classes the RLB Specialist Panel catalogues across regulators:

• Exposed Fabrication: the model commits to a specific wrong answer with confidence, then retracts when challenged with the regulator's text. The first-pass response is the one that enters a deliverable; the retraction happens only if a reviewer already knows the correct answer.

For compliance teams at corporate banking firms, the practical signal is that AI assistants with web search enabled remain prone to Inference Drift on event-confirmation questions and to Exposed Fabrication on scope-boundary questions. The first failure mode is dangerous because it looks responsive: the deliverable carries the AI's confirmation that something happened, but the specific load-bearing detail is missing. The second failure mode is dangerous because the first-pass answer is delivered with no signal that it is wrong, and the retraction only fires for a reviewer who already knows the right answer.

Both modes survive ordinary review and only surface against the regulator's primary text.

What the RLB Specialist Panel is doing about it

The Panel runs Specialist Panel direct questions and Specialist Panel application-style questions against frontier AI models on every reg-rooted workflow compliance teams at corporate banking firms actually use AI for. Each surfaced hallucination is bound to a verbatim regulator-issued anchor before publication, and each is recorded with a citation ID that traces the question, the AI's response, the verbatim source text, and the audience-specific operational consequence.

The Panel works directly with institutional readers, AI labs, and regulator-facing teams to feed back the patterns it records, so the same failure modes can be addressed at source rather than caught at the reviewer's desk. For the the CFTC swap dealer business conduct and documentation rulemaking, the Panel's substrate covers the December 2025 final rule, the January 2026 correction notice, the eCFR text of 17 CFR Part 23 Subpart H, and the chain of CFTC staff no-action letters that govern cross-border ITBC swap execution.

What Corporate-banking compliance teams should do

• Verify against primary substrate: where the deliverable depends on a specific appendix identity, staff no-action letter scope, or product boundary from § 23.431, do not rely on the AI's summary; open the eCFR, the Federal Register, or the CFTC staff letter and confirm the load-bearing detail.
• Treat event-confirmation as the start of a verification step, not the end: if the AI confirms that a correction notice was issued, that a staff letter exists, or that a requirement was eliminated, treat that as the cue to identify which appendix, which venues, or which products are actually in scope.
• Build a primary-source check into the sign-off workflow: for the written supervisory procedure update, compliance attestation, audit walkthrough documentation, or supervisory communication on swap dealer business conduct this audience produces, introduce a documented check on every appendix citation, venue-scope claim, and product-boundary statement before sign-off.
• Use the Panel's citation IDs as a reference index: when an AI answer matches a known Panel finding, the cited regulator text is already provided as the corrective anchor.

The Panel records the one hallucination above with citation ID RLB-H-US-CFTC-SWAP-DEALER-BUSINESS-CONDUCT-DOCUMENTATION-2025-Q004-Opus47 for direct reference in this audience's workflow.

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: 17 CFR Part 23, Subpart H, CFTC Swap Dealer Business Conduct Standards Documentation · Substrate documents: cftc-final-rule-2025-23953-fr-90fr61226.pdf · eCFR: ecfr.gov · CFTC: cftc.gov

Citation IDs referenced:

• RLB-H-US-CFTC-SWAP-DEALER-BUSINESS-CONDUCT-DOCUMENTATION-2025-Q004-Opus47