Executive Summary
The CPMI's October 2024 framework on harmonising application programming interfaces for cross-border payments sits across the surface that Financial Advisers in international jurisdictions touch in market briefings, client memos, and corridor-strategy advisory work. The 70-plus global fast payment systems landscape, the central-bank versus private operator mix, and the named jurisdictional implementation partners (such as the South African Reserve Bank on payment pre-validation) are the kind of factual detail that gets quoted into adviser-facing deliverables without independent verification.
In our research, AI tools issued responses on 2 questions on this regulation that, while phrased with authoritative confidence, misstated the regulator-published global FPS count and falsely declared a CPMI operational-split statistic unavailable. A Financial Adviser who uses AI to populate primary-source benchmark data on this regulation receives outputs that look correct, present cleanly in client materials, and contradict the primary record.
How AI gets this regulation wrong
The table below catalogues how AI tools erred when asked about this regulation in the contexts Financial Advisers most often encounter it. Failures cluster around AI substituting inference for retrieval, typically reading a category name, a publication abstract, or an adjacent CPMI speech and then generating structural detail that has no support in the regulator's own primary text. The AI does not flag the substitution; the inferred detail is presented in the same register as retrieved fact.
| AI's Failure Mode | Count | Affected findings |
|---|---|---|
| 0 |
What that means for your practice
The errors documented here translate into the wrong-deliverable risk category for Financial Advisers: a client memo, market briefing, or corridor strategy document that includes fabricated CPMI statistics or asserts non-existent data gaps. The downstream cost is reputational, clients act on the wrong figures, internal review reveals the error after circulation, or a regulator referencing the same CPMI primary source surfaces the discrepancy in supervisory dialogue.
| Risk Impact | Count | Affected findings |
|---|---|---|
| 0 |
When this affects your work
Financial Advisers reach for AI on this regulation most often when assembling market briefings or corridor strategy memos that depend on CPMI benchmark statistics: how many fast payment systems are operational globally, which jurisdictions are linking cross-border, what the ownership split between central banks and private operators looks like across the universe. The benchmark data is published in CPMI's own primary sources (the November 2023 Tara Rice speech is the canonical reference for current-state FPS statistics), but those sources sit across multiple documents that AI tools are erratic in retrieving in full.
The risk is that a briefing built on an AI-sourced figure under-reports the global FPS universe, quoting the 2025 monitoring survey sample of 57 systems instead of the regulator-stated 70+, or asserts a data gap that does not exist. The client acts on a smaller landscape, the corridor strategy targets the wrong subset of jurisdictions, or the briefing simply omits a CPMI-published statistic the AI failed to find. None of these errors are dramatic on the surface; all of them are recoverable to the primary CPMI source within a single browser session.
The cost is the briefing that went out before that verification happened.
The findings at a glance
The table below summarises each finding from our research on this regulation for Financial Advisers, including a short title, the type of AI failure observed, and the citation reference for the finding card below.
| # | Finding title | Type | Citation ID |
|---|---|---|---|
| 1 | Global FPS count compressed to 57 from 70+ | Hallucination | RLB-F-INT-BIS-CPMI-API-HARMONISATION-CROSS-BORDER-2024-Q010 |
| 2 | Central-bank vs private FPS split falsely declared unavailable | Hallucination | RLB-F-INT-BIS-CPMI-API-HARMONISATION-CROSS-BORDER-2024-Q010 |
Aggregate impact
The two findings on this regulation cluster on the same underlying defect: AI tools substituted a smaller-scope source (a monitoring-survey sample) for an authoritative source (a CPMI speech documenting the universe figure), or denied that a published statistic exists when it does. For Financial Advisers building market briefings, the systemic exposure is that AI-sourced benchmark data on this regulation is not reliable for client-facing work. The errors here are recoverable to the primary CPMI publications within a single verification step; the risk is the briefing that ships before the verification happens.
The broader pattern across CPMI primary sources is that the regulator documents statistics in speeches and briefs that AI retrieval handles erratically, while the monitoring-survey outputs and aggregator coverage that AI tools handle better are precisely the documents that report sample-level figures rather than universe-level figures. A Financial Adviser who wants to rely on AI for CPMI benchmark data needs to know that this asymmetry exists, because the AI does not signal it: a sample number presented as a universe number reads the same as a correctly retrieved figure in the response.
What you should do
Treat AI-sourced CPMI benchmark statistics as a flag to verify, not as the figure to quote. The CPMI publication feed at bis.org/cpmi carries the speeches and briefs where current-state global FPS data is published, and the Tara Rice November 2023 speech in particular is the canonical primary source for the 70+ universe figure and the central-bank versus private operator split. A standing practice of confirming any AI-supplied CPMI statistic against the primary publication before it enters a briefing avoids the universe-versus-sample substitution that the AI does not flag.
Where AI tools do add value on this regulation is at the framing and orientation level: summarising why API harmonisation matters for cross-border payments, sketching the high-level structure of the 10 recommendations, or drafting an outline for a client briefing that will subsequently be populated with primary-source verified statistics. The risk concentrates where the AI moves from framing to quantitative claim or named institutional partnership. A simple verification protocol, every AI-sourced number on this regulation gets a primary-source check, is sufficient to keep adviser-facing deliverables reliable.
How RLB Can Help
RegLeg's Hallucination Research is built for exactly the deliverable workflow Financial Advisers run: market briefings, client memos, and corridor strategy notes where AI is used to accelerate the data-assembly stage. The published research catalogues where AI tools have produced confident but wrong answers on regulatory benchmark statistics, named institutional partnerships, and primary-source data availability claims. Knowing those failure patterns in advance lets an adviser apply targeted verification, checking the specific categories of claim that the research shows are unreliable, rather than blanket scepticism that would make AI assistance impractical.
For advisory practices that want to systematise this, RegLeg can run a confidential review of the firm's current AI-use practices in adviser workflows and return a prioritised list of where verification controls are missing or under-specified. The review is calibrated to adviser realities: time pressure, client-facing deliverable cycles, and the difference between framing AI use (low risk) and quantitative claim AI use (higher risk requiring primary-source verification).
RegLeg also develops CPD-aligned training content tailored to advisory contexts, scenario-based, grounded in real documented failures from the research, and designed for experienced advisers who do not need the AI 101 but do need a defensible protocol for using AI assistance on regulatory benchmark and statistical content.