Hallucination findings aggregated by (sub-sector × department) per jurisdiction. Each case study covers one cell and draws on all regulations relevant to it.
Last updated 14 Jun 2026 · Full ledger: Hallucination Register · Sibling: Practitioners · AI Labs
AI tools consistently misstate, drop, or invert critical qualifiers in Consumer Duty rules — including the 'reasonably believes' carve-out, the charity...
AI tools tested on the FCA's Consumer Duty framework produced errors on every question, including substituting a legally distinct financial concept for a...
AI tools consistently misstate the scope, rule text, and supervisory landscape of Consumer Duty (PS22/9), creating a systemic risk that Life Insurance firms...
AI tools tested on Consumer Duty obligations cited a non-existent binding rule in PRIN 2A and elevated non-binding FCA guidance on consumer testing to...
AI assistants consistently misstate whether Consumer Duty obligations are binding rules or guidance recommendations — a distinction that directly determines...
AI tools consulted by UK Retail Banking Legal teams on Consumer Duty scope and legal basis produce answers that are partially correct but contain material...
AI tools fabricated a specific binding rule requiring communications testing under Consumer Duty — complete with a false Handbook rule reference — and...
AI tools consistently misrepresent the FCA's Consumer Duty fair value standard by inflating the obligation to quantify non-monetary benefits — an error...
AI tools tested on Consumer Duty harm-prevention obligations replaced the FCA's single qualifying condition with a fabricated multi-condition test, creating...
AI tools consistently told Corporate Banking Compliance teams that the CPMI-IOSCO 2016 Cyber Resilience Guidance is current and definitionally aligned with...
AI tools characterise the 2016 CPMI-IOSCO Cyber Resilience Guidance as already containing the granular incident response and recovery practices that were...
AI tools asked about the CPMI-IOSCO 2016 Cyber Resilience Guidance consistently overclaim both what the document explicitly cites and how granular its...
AI tools consistently overstate the operational completeness of the 2016 CPMI-IOSCO Cyber Resilience Guidance, attributing detailed incident response and...
AI tools consistently overclaim the scope and specificity of the CPMI-IOSCO 2016 Cyber Resilience Guidance, fabricating explicit framework citations and...
AI tools consistently misrepresent both the operational detail of the CPMI-IOSCO 2016 Cyber Resilience Guidance and its current regulatory status — with...
AI tools confidently characterise the 2016 CPMI-IOSCO Cyber Resilience Guidance as containing detailed operational expectations for incident response and...
AI tools systematically overclaim the depth and external alignment of the CPMI-IOSCO 2016 Cyber Resilience Guidance, creating a risk that Payment...
When a Retail Banking firm's Compliance team asks AI tools whether the CPMI-IOSCO 2016 Cyber Resilience Guidance is still operative, multiple AI tools...
AI tools told Compliance teams that the CPMI-IOSCO 2016 Cyber Resilience Guidance remains the operative international standard — unaware that CPMI-IOSCO had...
AI tools confidently assert that the CPMI-IOSCO 2016 Cyber Resilience Guidance formally cites the NIST Cybersecurity Framework — a claim that cannot be...
AI tools fabricated a non-existent notice designation for Financial Holding Companies when asked about the scope of MAS Notice 637, a type of error that...
AI assistants tested on MAS capital adequacy rules fabricated a specific notice designation for financial holding companies — a plausible but unverifiable...
AI tools consulting on MAS Notice 637 annex content have been found to substitute Basel III structural inference for the actual regulatory text, creating a...
AI tools asked about MAS Notice 637 fabricated a regulatory instrument designation for financial holding companies, mischaracterised a document-marking...
AI tools asked about the structure of MAS Notice 637 produced characterisations of specific annexes that appeared to be inferred from general Basel III...
AI assistants mischaracterised the scope and content of specific annexes within MAS Notice 637, producing confident but unverifiable or contradicted...
AI tools fabricated a plausible-sounding but non-existent MAS notice designation for financial holding companies, a type of error that would propagate...
AI tools mischaracterised the structural content of MAS Notice 637 by presenting an unverified inference about a specific Division of Part VI as a...
AI tools consulting on MAS Notice 637 fabricated a non-existent notice designation for financial holding companies and misidentified a precise...
AI tools tested on MAS Notice 637 annex-level queries produced responses based on inferred Basel III framework structure rather than the actual regulation...
AI tools queried on the structure of MAS Notice 637 inferred divisional content from surrounding context rather than retrieving it, producing responses that...