Enumeration Collapse on a CFTC margin rule: two frontier AI models rewrote the Regulation 1.44 currency deadline tiers in ways that, if propagated into an investment bank's transaction-side documentation, would mis-state a counterparty's compliance posture and commit the desk to a standard the CFTC has not set.

The exposure sits at the intersection of AI-assisted legal review and the volume of margin-touching documentation an investment bank produces on FCM counterparties, multi-currency client accounts, and cross-jurisdictional client funds.

The pattern in one line

Frontier AI models, asked for an operational map of CFTC Regulation 1.44's currency deadline tiers, returned two structurally wrong outputs: one collapsed three regulatory tiers into two, the other added a noon Eastern Time cutoff that does not exist in the rule. An investment bank legal review built on either output would document the wrong standard inside diligence files, client disclosures, MD briefings, and counterparty risk memos.

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. For Regulation 1.44, the Panel set a deliverable-style operational brief: prepare guidance for a margin team configuring system parameters for Regulation 1.44 separate account compliance, specifying which currencies require same-day collection, which receive extended deadlines, and the exact timing requirements for each currency category. Two frontier AI subjects answered with web search enabled, mirroring how transaction-side legal teams actually use AI assistants on a finalised rule with deal timelines pressing on review windows.

The Panel binds each AI finding to verbatim regulator-issued source text held as primary substrate, in this case the eCFR text of 17 CFR Section 1.44 as of June 2026.

What the models got wrong

Claude Opus 4.7, with web search active, wrote: "For margin called in a fiat currency other than USD, the deadline may be extended by up to one additional business day... T+1 should be the default presumption... For all other non-USD currencies, default to same-day (T)." The structural error has two arms. Appendix A currencies, the ten currencies enumerated in the rule (AUD, CNY, HKD, HUF, ILS, JPY, NZD, SGD, ZAR, TRY), were assigned T+1.

Section 1.44(f)(2) sets T+2, "the end of the second business day after the day on which the margin call is issued." Every remaining non-USD, non-Canadian-dollar currency was pushed to same-day. Section 1.44(f)(3) grants those currencies T+1, end of the business day after the day on which the call was issued. Three tiers compressed to two. Citation ID RLB-H-US-CFTC-FCM-MARGIN-ADEQUACY-SEPARATE-ACCOUNTS-REG-1-44-Q001-Opus47 carries the full audit trail.

Claude Sonnet 4.6, given the same brief, wrote: "TIER 1 EXTENSION, OTHER NON-USD/CAD FIAT CURRENCIES (T+1, 12:00 p.m. ET) Deadline: 12:00 p.m. ET on the FIRST U.S. business day after the business day on which the margin call was issued." Section 1.44(f)(3) sets no intraday cutoff. The verbatim text reads: "no later than the end of the business day after the day on which the margin call is issued." The "12:00 p.m. ET" figure does not appear in the rule or Appendix A.

A separate Specialist Panel re-probe on a related Sonnet 4.6 question surfaced a fabricated citation: the model supported a two-tier reading with a URL on the website of an international law firm, and the Panel verified the cited page does not exist. Citation ID RLB-H-US-CFTC-FCM-MARGIN-ADEQUACY-SEPARATE-ACCOUNTS-REG-1-44-Q001-Sonnet46 carries the full audit trail.

Why this matters for investment banking legal teams

The exposure follows the bank's documentation chain. The transaction-side diligence summary that names an FCM counterparty's margin call timing posture is the first deliverable an AI-assisted reviewer will produce, and a summary built off the Opus 4.7 reconstruction will state that the FCM is late on Appendix A margin received on T+2, which is the regulatory deadline. That mis-statement enters the diligence file, the deal memo, and the counterparty risk register.

The client disclosure language validated against the Sonnet 4.6 output will assert a noon Eastern Time cutoff as a regulatory obligation; that language travels into the client agreement and into the bank's documented standard of conduct on multi-currency margin processing. A 2-page MD briefing summarising either error becomes the desk's operating view of the rule, which propagates into trader training, into the FCM relationship review, and into examination response materials. Each derived deliverable inherits the same error, and opposing counsel in a margin dispute will read the bank's documented position before the rule.

The regulator's actual position

Section 1.44(f) sets three tiers, defined by appendix membership rather than by intuitive currency grouping. Tier 1, U.S. and Canadian dollars, close of the Fedwire Funds Service on the same business day the margin call is issued. Tier 2, the ten currencies listed in Appendix A, end of the second business day after the day on which the margin call is issued. Tier 3, all other fiat currencies, end of the business day after the day on which the margin call is issued. No intraday cutoff appears in any of the three tiers.

The Tier 2 membership list is regulator-defined, not derivable.

What this tells us about AI for investment banking legal teams

The taxonomy lens for the bank is Enumeration Collapse: the substitution of a plausible structure for the regulation's actual enumeration, presented as a usable deliverable with no surface signal to the reader. Both Regulation 1.44 outputs were internally consistent, formatted as desk-usable guidance, and carried no hedging. The error is not catchable from the output itself. For investment bank legal teams, the operational implication is that any AI-assisted regulatory review used in transaction-side documentation must be verified against the primary text before the documentation is signed out.

The volume of AI-assisted review in deal workflows magnifies the exposure: an unverified Regulation 1.44 paragraph in a single diligence summary will be cited in every derived deliverable downstream of it.

What the RLB Specialist Panel is doing about it

The RLB Specialist Panel records both findings under the immutable citation IDs above and binds each to the verbatim Section 1.44(f) text it contradicts. The Panel records the failure mode classifications (outdated for the Opus 4.7 finding, inference_drift for the Sonnet 4.6 finding) against the eCFR primary substrate document. The Panel sets out the methodology categories, Enumeration Collapse and Fabricated Intraday Cutoff, so investment bank legal teams and AI labs can identify the same patterns in their own workflows.

Partnership inquiries from banks that want to test their own AI-assisted regulatory review against the same probe family, and from AI labs that want to test their models against the same brief, are received at the Specialist Panel level.

What investment banking legal teams should do

Documentation-chain actions, in order of bank-level exposure:

• Treat every AI-assisted Regulation 1.44 paragraph in diligence summaries, client disclosures, MD briefings, and counterparty risk memos as draft until verified against 17 CFR Section 1.44(f) and Appendix A directly.
• Flag intraday Eastern Time cutoffs in any AI output on the T+1 tier as a diagnostic of reconstructed text. No such cutoff exists in the rule; its presence in an AI output is a verification trigger.
• Name-check every currency in an AI-generated Tier 2 list against the Appendix A enumeration before the list enters any documented deliverable.
• Verify every URL the AI cites as authority. Fabricated URLs to known law-firm domains have been observed on this rule.
• Inventory the bank's existing Regulation 1.44 documentation that was generated with AI assistance since the rule was finalised and re-verify each entry against the primary text. Errors propagate forward from the first unverified deliverable.
• For transaction-side reviews, route AI-assisted drafts through a primary-text verification step that is logged in the deal file, so any later correction is traceable to its source.

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 § 1.44, Risk-based Capital Requirements and Margin Adequacy for Separate Accounts · Substrate documents: R2-REGULATION-17CFR_1_44_eCFR_asof_2026-06-04.pdf · eCFR: ecfr.gov · CFTC: cftc.gov

Citation IDs referenced:

• RLB-H-US-CFTC-FCM-MARGIN-ADEQUACY-SEPARATE-ACCOUNTS-REG-1-44-Q001-Opus47