← RegLegBrief Findings by regulator All AI Labs whitepapers Hallucination Register Methodology Right of reply
AI Labs · White Paper · BIS-CPMI · INT · substrate v1

Two adoption rates. One blended number.
Both FPS and RTGS figures wrong.

Claude Opus 4.7, with web search active, collapsed the CPMI's separately reported ISO 20022 adoption rates for faster payment systems and RTGS systems into a single blended claim. The 2026 CPMI harmonisation update keeps these figures disaggregated. The model's blend exists in neither the FPS figure nor the RTGS figure.

📅 Published 7 Jun 2026 ⚙️ Methodology v2.3 📡 CPMI ISO 20022 2026
<Document> CPMI ISO 20022 Harmonisation 2026 </Document> · FPS_adoption: [disaggregated] · RTGS_adoption: [disaggregated] · Model_output: [blended] ✗
📰Read the public briefing for this regulation
The two distinct channels

FPS and RTGS are separate adoption tracks in the CPMI data

The CPMI's 2026 ISO 20022 harmonisation update tracks adoption across two structurally different payment infrastructure channels: faster payment systems (FPS) and real-time gross settlement systems (RTGS). These are not subcategories of the same metric. they serve different transaction types, operate on different timescales, and have different adoption economics. The CPMI reports their ISO 20022 migration status separately, because the figures are different and the policy questions around each channel are different.

Channel 1: FPS
Faster Payment Systems
Consumer and retail payment rails. Many were built natively on ISO 20022 or migrated early. High adoption rate. CPMI tracks separately because FPS architecture differs from RTGS.
Channel 2: RTGS
Real-Time Gross Settlement
Large-value interbank payment systems. Legacy MT-format infrastructure more prevalent. Migration timeline and adoption economics differ from FPS. CPMI tracks separately.
Hallucination Type: Numeric Conflation across Disaggregated Subcategories
Claude Opus 4.7 collapsed the two separately reported adoption figures into a single blended claim. The resulting number does not correspond to either the FPS adoption rate or the RTGS adoption rate recorded in the CPMI 2026 harmonisation update.
Attribution failures

Wrong numbers attributed with confidence to the CPMI document

The compounding problem here is attribution. Claude Opus 4.7 did not hedge or caveat the blended adoption figure. it attributed it directly to the CPMI harmonisation document with the same confidence it would use for a figure that actually appears in the text. A product builder, researcher, or compliance professional reading the model output has no signal that the specific number is a blend rather than a direct quotation.

Attribution drift of this kind, where a fabricated or conflated figure is attributed to an authentic source document with citation-level confidence, is a distinct failure class from simple numeric error. It defeats the primary QC mechanism available to a downstream user, which is checking the source.

ISO 20022 ADOPTION — CPMI DATA vs MODEL OUTPUT CPMI 2026 (actual) FPS adoption Rate A % RTGS adoption Rate B % ← two separate figures Model output (Opus 4.7) Blended single figure: "ISO 20022 adoption: X%" Attributed to CPMI document · does not appear in either FPS or RTGS subcategory ✗
The CPMI 2026 update reports FPS and RTGS ISO 20022 adoption as separate figures. The model collapsed both into a single blended number attributed with confidence to the CPMI source.
What this means for builders and researchers

The operational failure surface

Teams building on CPMI ISO 20022 data for cross-border payment product development, compliance tooling, or policy research need the disaggregated figures. A FPS product team asking a model about FPS adoption rates to benchmark a migration roadmap, and receiving a blended RTGS-contaminated figure, is working from a wrong input. A researcher citing the model's figure in a policy brief on payment system modernisation is publishing a number that is traceable to no CPMI document.

Model attribution (wrong)
Blended single adoption figure attributed to the CPMI 2026 harmonisation document. Does not appear in the document as a single figure.
Correct position
CPMI 2026 reports FPS adoption and RTGS adoption as separate disaggregated figures. Neither subcategory figure matches the model's blended output.
// Regulation context
CPMI ISO 20022 Harmonisation Updated 2026 — BIS Committee on Payments and Market Infrastructures. Cross-border payment API harmonisation and ISO 20022 adoption tracking across FPS and RTGS channels. Source: CPMI-ISO-20022-HARMONISATION-UPDATED-2026 →

Hallucination Register: reglegbrief.com/hallucination-register/ · Right of Reply: reglegbrief.com/right-of-reply/