This is the consolidated view of findings. Click the Citation IDs or 'see details →' on any item for the full details for each finding.
When a lawyer asks AI tools about the oversight expectations applicable to critical service providers under PFMI Annex F and the relationship to the companion assessment methodology document, the AI produced an incorrect document identification — misassigning a document number to a different CPMI publication. A lawyer who relies on this to structure an engagement or frame a regulatory opinion will be working from a misidentified source, potentially scoping the wrong obligations or citing the wrong standard in advice to a client.
In regulatory examinations or cross-border compliance reviews where the critical service provider oversight framework is live, the practical consequence is advice based on the wrong document.
Asked about the specific governance requirements in PFMI Principle 2 — including whether establishing a risk committee is a conditional recommendation or a mandatory requirement — an AI tool fabricated a specific key consideration number and quoted committee governance language it had not verified, citing a third-party document as supporting authority when the source did not directly verify the claim.
A lawyer advising an FMI board on its governance obligations who takes this output at face value risks providing an incorrect characterisation of the mandatory/discretionary distinction, which is a live issue in board governance reviews and regulatory examinations across PFMI-implementing jurisdictions. The Pretextual citation compounds the risk: the AI's source does not establish what the AI claims, but a quick review of the cited URL could be mistaken for verification.
Two AI tools tested on the August 2016 CPMI-IOSCO consultative report on CCP resilience and recovery both correctly acknowledged they could not retrieve verbatim text, specific thresholds, or paragraph cross-references from the document. For a lawyer advising on CCP recovery planning, stress-testing obligations, or loss allocation rules in any PFMI-implementing jurisdiction, this means AI tools cannot supply the actual regulatory language that would need to appear in a legal work product — the source document must be obtained and read directly.
The Pretextual citation label on the BIS URL indicates the AI directed the lawyer to the document without being able to verify its content.
The November 2025 CPMI-IOSCO Level 3 assessment on general business risks — which includes findings on FMI compliance with the six-month liquid net assets standard under PFMI Principle 15 — was confirmed by both AI tools to be beyond their training data and inaccessible in verbatim form.
For a lawyer advising on capital adequacy or liquidity risk governance for an FMI or its participants, the inability of AI tools to surface findings from the most recent Level 3 assessment means that advice based on AI research will be structurally out of date on a document that was published after the AI's knowledge cutoff. The Pretextual citations indicate the AI was aware the document existed but could not supply its content.
Both AI tools tested on the July 2022 BIS press release announcing the CPMI-IOSCO stablecoin guidance publication correctly declined to provide verbatim text, acknowledging they could not guarantee accuracy without retrieving the page directly. For a lawyer advising on the application of PFMI standards to stablecoin arrangements — an area of active regulatory development — the inability of AI tools to surface the specific content of regulatory communications means any AI-assisted research on this topic needs to be treated as directional only, with the source materials retrieved and reviewed before use in legal advice.
Both AI tools correctly declined to supply verbatim text from the IOSCO co-published version of the PFMI (IOSCOPD377), noting that the binary PDF was not accessible at paragraph level. In practice the IOSCO version carries the same substantive content as the BIS publication, but for lawyers advising IOSCO-regulated entities or filing documents with IOSCO-member regulators who reference the IOSCO-published version as the applicable instrument, the inability of AI tools to retrieve its precise text creates a gap that must be filled by direct document review.
Advice or submissions that paraphrase rather than accurately quote the applicable instrument carry regulatory and professional risk.
Both AI tools correctly declined to provide verbatim content from the IOSCO co-published version of the CPMI-IOSCO disclosure framework and assessment methodology (IOSCOPD396 / d106), noting the binary PDF was not available to them at paragraph level. For lawyers supporting FMI self-assessments or supervisory assessments, the disclosure framework and assessment methodology is the operational document that translates the PFMI Principles into assessment criteria — and the inability of AI tools to supply its precise wording means any AI-assisted work in this area must be underpinned by direct access to the published document rather than AI-generated paraphrase.