Executive Summary
Compliance teams at Corporate Banking firms in the United States face a specific AI reliability problem when working with the CFTC's December 2025 final rule revising business conduct and swap documentation requirements for swap dealers and major swap participants. AI tools we tested failed on questions about the scope of the pre-trade mid-market mark (PTMMM) disclosure requirement and what the rule's elimination of that provision actually covered. Out of the questions tested against this regulation, AI tools produced at least one aggregated hallucination, confidently asserting a product-scope claim that was factually wrong, and then retracting it under challenge.
The failure pattern is product-scope fabrication: the AI overstated which instrument categories had been freed from PTMMM obligations, asserting relief for cleared instruments that were never inside the requirement's scope to begin with. For a swap dealer's Compliance function, that kind of error is operationally dangerous precisely because it sounds technically plausible, it reads like a correct reading of "eliminated in its entirety."
How AI gets this regulation wrong
The AI failures on this regulation centre on product-scope fabrication, where AI tools answered a precise question about instrument coverage with confident, fluent misinformation, then retracted under pressure. The breakdown is not a vague misreading of an ambiguous rule; it is a specific factual error about which instrument categories were ever within §23.431(a)(3)'s scope, dressed up in the regulatory vocabulary your team uses every day. The table below maps the failure modes as they actually appeared.
| AI's Failure Mode | Count | Affected findings |
|---|---|---|
| Exposed Fabrication | 1 | Finding#1 |
What that means for your team
The risk this error creates for your function sits squarely in the regulatory enforcement category, not because the CFTC is likely to examine your internal research notes, but because the misunderstanding travels downstream into policies, training scripts, and desk-level guidance that your examiners will eventually read. A Compliance team that has internalized a wrong product-scope boundary for a pre-trade disclosure requirement is producing defective controls from day one of the new rule's effective date. The table below maps where the enforcement exposure concentrates.
| Risk Impact | Count | Affected findings |
|---|---|---|
| Regulatory enforcement | 1 | Finding#1 |
When this affects your department
Swap dealer Compliance teams reach for AI tools most heavily during the gap between a final rule's publication and its effective date, exactly the window when this CFTC rule is live. The immediate workload is dense: updating pre-trade disclosure policies, rewriting the desk's counterparty communication templates, briefing front-office staff on what changed and what didn't, and confirming to internal audit that the firm's §23.431 controls are correctly scoped. AI tools are tempting here because the regulation is granular, the preamble is long, and junior team members need fast, authoritative-sounding answers.
The specific question where AI tools failed, which instrument categories were actually subject to the PTMMM requirement before the rule, and whether "eliminated in its entirety" means the obligation is gone for every swap type, is exactly the kind of scoping question that lands in a Compliance analyst's inbox on day one of implementation. If a junior analyst asks an AI tool and takes the answer at face value, the firm's desk-level policy ends up stating that cleared credit default swaps (index and single-name) are now exempt from PTMMM, framing an exemption where no requirement ever existed.
That error embeds itself in training materials and counterparty-disclosure checklists before anyone with institutional memory of the pre-rule regime has reviewed the draft.
What's at stake is not just a documentation defect. If the Compliance function has incorrectly mapped product scope, internal audit and the first-line business may both inherit that mapping. When the CFTC's Division of Swap Dealer and Intermediary Oversight conducts its next examination, the firm's written policies and front-office attestation records may reflect a coherent but factually wrong account of which instruments the pre-trade disclosure framework covered, and what the 2025 amendments actually changed.
That creates examination findings and, in a worst case, a basis for enforcement action premised on a firm that cannot demonstrate accurate understanding of its own regulatory obligations.
The findings at a glance
The table below summarises the finding produced by AI tools on this regulation as tested against the Compliance × Corporate Banking lens in the United States, including the question area, the nature of the failure, and where the risk lands.
| # | Finding title | Type | Citation ID |
|---|---|---|---|
| 1 | PTMMM elimination scope, cleared vs. uncleared instrument coverage | Hallucination | RLB-F-US-CFTC-SWAP-DEALER-BUSINESS-CONDUCT-DOCUMENTATION-2025-Q004 |
Aggregate impact
The single finding from this regulation illustrates a failure pattern that is structurally more dangerous than a simple factual error: the AI gave a confident, technically fluent answer, and only retracted when directly challenged. That sequence, confident assertion, downstream use, late retraction, is the worst-case scenario for a Compliance team under time pressure. The initial answer sounded like exactly the kind of crisp product-scope analysis a senior reviewer would produce. The problem is that most implementation timelines do not include a round of adversarial follow-up questioning before policy drafts circulate.
The substance of the error is also precisely calibrated to mislead. The AI's claim that cleared CDS, index and single-name, are now exempt from PTMMM is not random noise. It is a logical extrapolation from "eliminated in its entirety," applied without the institutional knowledge that cleared instruments were never inside §23.431(a)(3)'s scope in the first place. That gap between the rule's plain language and its historical product scope is exactly where AI tools fail: they read current text accurately but reconstruct prior scope incorrectly, producing a delta narrative that is plausible but wrong.
For a Corporate Banking firm's swap dealer Compliance function, the aggregate risk is that this failure clusters on the most consequential implementation question, not a procedural detail, but the foundational scope question: which products does this requirement apply to, and which products does it not? Getting that wrong means the entire control architecture built on top of the answer is misaligned. Remediation once the error is discovered, redrafting policies, retraining desk staff, issuing corrected guidance, is expensive. Remediation after an examination finding is more expensive still.
What your team should do
The default position for your team on this regulation should be: AI tools are useful for drafting and summarising, but not for authoritative product-scope determinations on recently amended rules. The PTMMM error is a direct consequence of using an AI tool to answer a question that requires two things simultaneously: accurate reading of new rule text AND accurate knowledge of the prior regulatory scope it amended. AI tools tend to handle the first acceptably and the second poorly, particularly for rules that reorganised or restructured existing provisions rather than creating entirely new ones.
In practice, the safeguard is procedural rather than technical. Scope questions, which instruments are in, which are out, what changed and what did not, should be answered by reference to the rule text, the CFTC's preamble discussion, and your firm's own pre-rule compliance documentation, not by AI summary. AI tools are reliable for tasks like reformatting regulatory text into policy language, flagging sections of the rule that address a topic you have identified, or drafting counterparty disclosure templates once the scope has been confirmed by a qualified reviewer.
They are not reliable for reconstructing the historical scope of a provision that has just been amended.
The specific red flag to watch for on this rule is any AI response that describes cleared instruments as having gained a new exemption or relief from PTMMM as a result of the December 2025 amendments. Cleared swaps were outside §23.431(a)(3) before the rule and remain outside the reorganised provision after it, the amendment changed the structure and location of the pre-trade price and compensation disclosures for instruments that were always in scope, not the product universe itself.
If your team's draft policies or training materials reference cleared CDS as beneficiaries of the rule change, that framing should be corrected before the documents circulate beyond Compliance.
How RLB Can Help
RegLeg's published Hallucination Research is available as a free pre-flight check before your team relies on AI output for any regulatory question. Before an AI-assisted BSA/AML opinion reaches a compliance memo, before a sanctions-screening gap analysis gets cited in a board paper, or before a CRA stress-testing narrative goes to examiners, the research lets you see, regulation by regulation, exactly where AI tools have fabricated statutory text, inverted regulatory scope, or confabulated agency guidance that does not exist. That is a concrete quality gate, not a theoretical one, and it costs your team nothing to run.
Where the free research ends, RegLeg works directly with corporate banking compliance functions on bespoke regulator deep-dives. That means mapping your specific AI-supported workflows, model-risk attestation, OCC/Fed exam preparation, Reg W affiliate-transaction monitoring, DFAST narrative drafting, against the failure-mode catalogue to produce a prioritised heat map of hallucination exposure by workflow and regulatory surface. The output is scoped to what your compliance team actually does, not a generic enterprise AI-risk framework, and it is built to sit alongside your existing Model Risk Management documentation rather than replace it.
For teams that have already deployed AI tools internally, RegLeg offers a confidential review of your firm's AI-use policy against our failure-mode catalogue, with prioritised remediation recommendations ranked by regulatory severity and examiner visibility. We also develop training material and CPD-aligned content calibrated for compliance professionals who do not need the 101 but do need to make credible, defensible decisions about where AI assistance is and is not appropriate in a heavily examined environment.
The goal is to give your team a grounded, evidence-based position on AI reliability, one you can put in front of a regulator if you have to.