AI Hallucination ResearchAudiencesSectorsInternational / MultilateralPayment InstitutionsRisk › Promoting the Harmonisation of Application Programming Interfaces to Enhance Cross-Border Payments: Recommendations and Toolkit
Payment Institutions × Risk — International / Multilateral · updated 2026-06-04 · methodology v2.3
Share / Print Twitter LinkedIn Email

AI on CPMI API Harmonisation Recommendations for Risk teams at Payment Institutions firms in international jurisdictions

This is the consolidated view of findings. Click the Citation IDs or 'see details →' on any item for the full details for each finding.

  1. AI fabricates recommendation-level stakeholder attribution
    RLB-F-INT-BIS-CPMI-API-HARMONISATION-CROSS-BORDER-2024-Q008

    When a Risk team member uses an AI-generated stakeholder matrix to map which of the 10 CPMI API harmonisation recommendations create obligations for the firm versus for payment system operators or standards bodies, a fabricated attribution table routes the firm's compliance scope analysis in the wrong direction from the outset — misidentifying whether a gap is the firm's to close or a systemic issue to monitor.

    For a Payment Institution operating across multiple international jurisdictions, that scoping error is not contained to one market: it propagates into regulatory obligation registers, product-launch risk assessments, and board risk reports across all in-scope jurisdictions simultaneously. The AI's willingness to retract when challenged provides no protection if the review workflow does not include a challenge step; and given that the full recommendation-level detail is in the inaccessible PDF, a reviewer without direct access to primary sources has no independent basis to flag the error.

    see details →
  2. AI misstates CPMI fast payment system landscape data
    RLB-F-INT-BIS-CPMI-API-HARMONISATION-CROSS-BORDER-2024-Q010

    A market briefing or board paper that cites a global fast payment system count of 57 — the survey-sample figure AI tools conflate with the global universe — rather than the authoritative 70+ operational systems understates the landscape the firm is navigating and the systemic risk it is exposed to across correspondent relationships and settlement chains. Separately, suppressing the CPMI-verified operator composition data (40% central banks, 35% private entities) removes a structurally important variable from counterparty risk and regulatory-relationship analysis.

    For Risk teams at Payment Institutions, both errors corrupt the baseline data that underpins cross-border product risk appetite, new-market entry assessments, and strategic regulatory submissions — and because the AI's initial response on both points read as authoritative, the errors are unlikely to trigger a secondary source check without an explicit verification protocol.

    see details →