Executive Summary
The December 2025 CFTC final rule revising business conduct and swap documentation requirements for swap dealers and major swap participants is an active compliance target — and AI assistants tested on it produced hallucinations across three distinct questions, each with direct professional liability exposure for counsel advising swap dealers. The failures are not fringe errors: they cover the corrective notice that reinstated a suitability guidance appendix, the jurisdictional scope of a cross-border no-action letter, and the precise product-scope of a pre-trade disclosure requirement that the rule repealed.
In every case, the AI answered confidently — and in two of three cases, only acknowledged the error when directly pushed back on. For a Lawyers drafting compliance opinions, advising on documentation workflows, or scoping regulatory sign-off obligations under this rule, any one of these failures would produce defective advice on a live, in-force rulemaking.
How AI gets this regulation wrong
Across this regulation, AI tools failed in two distinct patterns: overstating what the rulemaking actually accomplished, and substituting familiar structural templates for the actual terms of novel cross-border guidance. What makes both patterns dangerous is the delivery — confident, well-structured answers that required direct challenge to dislodge, and in one case still omitted the specific identity of the regulatory artifact at issue even when the general answer was correct.
| AI's Failure Mode | Count | Affected findings |
|---|---|---|
| Exposed Fabrication | 2 | Finding#2 · Finding#3 |
| Inference Drift | 1 | Finding#1 |
What that means for your practice
All three findings concentrate in the same risk bucket — liability and professional indemnity exposure — because they each arise in the precise verification tasks where swap counsel is most likely to reach for AI: confirming what a correction notice reinstated, establishing the jurisdictional scope of a no-action letter before advising on compliance posture, and determining whether a prior disclosure obligation still attaches to a client's specific product mix. These are not research tasks requiring deep synthesis; they are targeted lookups where a single wrong fact produces a wrong opinion.
| Risk Impact | Count | Affected findings |
|---|---|---|
| Liability / PI exposure | 3 | Finding#1 · Finding#2 · Finding#3 |
When this affects Lawyers
Swap counsel reaches for AI tools at predictable inflection points when a rule like this one goes final: scoping which existing obligations survive, which are modified, and which were eliminated; confirming whether correction notices or no-action letters issued after the final rule affect the compliance timeline; and verifying whether cross-border guidance — staff letters, CFTC-FCA coordination instruments — expands or contracts the rule's domestic scope. Each of these is a targeted lookup, not a research project, and that's exactly why the AI failure mode is so acute here.
The practitioner isn't asking for an essay; they're asking for a specific fact to anchor an opinion or a client memo, and a confident wrong fact is worse than no answer at all.
The liability surface is sharpest in two workflow moments. The first is drafting compliance opinions or regulatory sign-off letters for swap dealers implementing the rule — if counsel misstates the product scope of the eliminated PTMMM requirement (advising that it now no longer applies to cleared CDS, when cleared swaps were already outside scope before the rule), the opinion overstates the change and may cause the client to misalign their disclosure infrastructure.
The second is cross-border advisory work: counsel advising a swap dealer on ITBC swap execution across UK venues needs CFTC Staff Letter 25-49's exact scope, not a plausible-sounding answer that maps the letter onto the domestic SEF/DCM template. An opinion premised on the wrong jurisdictional scope is simply wrong, and it's the kind of wrong that surfaces in an exam or enforcement inquiry.
Training and onboarding contexts carry a quieter but durable risk. When a firm's swap legal team uses AI to brief a new hire or produce a training module on this rule, a hallucination embedded in that material becomes the baseline understanding for an entire class of junior attorneys. The January 2026 correction notice reinstating Appendix A to Subpart H — which contains the suitability guidance for swap dealers making recommendations to counterparties and special entities — is exactly the kind of technical housekeeping item that gets relegated to a footnote in AI-generated summaries.
Counsel who never registers that Appendix A was momentarily voided and then restored may advise on §23.434 and §23.440 obligations without appreciating that the applicable interpretive framework briefly had an uncertain status.
The findings at a glance
The three findings below cover the corrective notice, the UK trading venue no-action letter, and the PTMMM elimination scope — each a discrete point of failure with a distinct consequence for counsel's work product.
Aggregate impact
The three findings cluster around the same underlying dynamic: AI tools have a confident working model of what the December 2025 CFTC rule did structurally — they know it revised business conduct and documentation requirements, they know there was a January 2026 correction, they know the PTMMM provision was eliminated — but their factual detail at the citation level is unreliable. The gap between structural knowledge and citation-level precision is exactly the gap where swap counsel operates.
A memo that correctly frames the rule's purpose but misstates which appendix was temporarily voided, or which venues a no-action letter covers, fails the primary function of the work product.
The most telling pattern is the retraction behavior. In two of the three findings, the AI changed its answer — or acknowledged uncertainty — only when challenged directly. That means a practitioner who does not already know the correct answer has no reliable signal from the AI's initial response that anything is wrong. The answer looks like good research: it's structured, it's specific, it references the right statutory provisions.
The error is in a layer of factual specificity — the identity of the UK venues versus US SEFs, the scope of the PTMMM elimination across product types — that only reveals itself when checked against primary sources. A practitioner using AI to accelerate research rather than to replace it would catch this; a practitioner treating AI output as citable work product would not.
The cross-border dimension of the Staff Letter 25-49 finding is worth isolating as a systemic risk signal. AI tools appear to default to the domestic regulatory template when interpreting CFTC no-action letters, mapping ITBC swap treatment onto US SEFs and DCMs even when the letter specifically addresses UK MTFs and OTFs under FCA authorization. For any firm with material UK execution flow, this is a structural bias in the AI's interpretive defaults — not a one-off error — and it will reappear on any question that sits at the CFTC-FCA interface without explicit prompting to focus on the cross-border scope.
What your team should do
The default position for any work product touching this rule should be primary-source verification at the Federal Register level — not because AI is useless here, but because the failures identified are precisely in the facts that look most plausible in an AI-generated answer. An AI that correctly describes the rule's structure but misidentifies the appendix restored by a correction notice, or misreads the scope of a no-action letter, is more dangerous than an AI that produces obvious nonsense, because it requires domain knowledge to catch.
Before any memo, opinion, or compliance sign-off cites the January 2026 correction, the CFTC Staff Letter 25-49, or the scope of the PTMMM elimination, those specific facts should be verified against the published primary text — not against an AI summary of the primary text.
For cross-border work involving UK execution, the Staff Letter 25-49 finding is a concrete prompt to build explicit scope-checking into your intake process. When an AI provides analysis on a CFTC no-action letter or staff guidance, prompt it specifically to identify the trading venues or counterparty categories covered, and then verify that characterization against the letter's text. AI tools tested on this regulation defaulted to the domestic SEF/DCM framework even when the letter was specifically about UK venues — a pattern that will generalize to other cross-border CFTC instruments.
Do not assume the AI has correctly identified jurisdictional scope without checking.
AI tools are genuinely useful for this rule in tasks that don't depend on citation-level precision: drafting issue spotters for a new engagement, structuring a compliance timeline for a swap dealer implementing the rule, comparing the final rule's structure against the proposed rule to identify where CFTC made substantive changes.
Where AI accelerates good work here is in framing and organization, not in answering "what specifically does X provision say" or "which document governs Y obligation." Keep that distinction clear for your team, and build the habit of treating any AI answer that sounds authoritative on a specific provision or instrument as a hypothesis requiring primary-source confirmation before it goes into client work product.
How RLB Can Help
RegLeg's published Hallucination Research is available without a paywall — use it as a pre-flight check before relying on AI output on any regulatory question we've covered. If you're using AI tools to draft advice, check positions, or summarise requirements, the findings catalogue tells you specifically where those tools have been shown to hallucinate: wrong numerical thresholds, inverted obligations, misattributed scope, fabricated effective dates. That's the kind of error that lands in a client memo or a regulatory submission.
Knowing the documented failure pattern for a given rule before you run your AI query is a material risk-management step, not a nice-to-have.
For firms with multiple lawyers working the same regulatory portfolio, we run bespoke deep-dives scoped to your actual workload — the specific rules your practice group relies on, tested against the failure modes that matter for your drafting and advisory workflow. The output is a working reference your team can use at the matter level: here are the questions you should not delegate to AI tools on this regulation without independent verification, and here is what the tool got wrong when we tested it. That's a more defensible position than a generic AI-use caveat in your engagement terms.
We also produce training material and CPD-aligned content built around the failure-mode catalogue — designed for teams that need to get lawyers up to speed on where AI tools break down in regulatory practice, without sitting through vendor demonstrations of features. Separately, if your firm has an existing AI-use policy, we can run a confidential review against our failure-mode catalogue to identify gaps: obligations your policy doesn't address, failure categories your review workflow doesn't catch, and places where the policy's permitted-use boundaries are looser than the evidence warrants.