AI Hallucination ResearchRegulatorsMajor advanced economiesSGMASNOTICE-637-CAPITAL-ADEQUACY-BANKS-2025White paperDetail › Finding
AI Labs · published 2026-05-28 · methodology v2.1

Claude Sonnet 4.6 with web search

What the RLB Specialist Panel found
  • Question (paraphrased to protect IP): What does Annex 4D of the consolidated MAS Notice 637 cover, and what category of items does it explicitly exclude from its scope?
  • AI's response: "sets out the credit conversion factors (CCFs) for off-balance sheet items in the context of the leverage ratio framework … Note: some search summaries linked Annex 4D to the SA credit risk framework rather than the leverage ratio; the precise scope should be verified"
  • Regulator's text: Annex 4D of the consolidated MAS Notice 637 addresses a defined topic area within the notice's capital adequacy framework that the model's response does not correctly identify.
  • Why the AI went wrong: The model produced a specific content description for Annex 4D — credit conversion factors for off-balance sheet items in the leverage ratio context — while simultaneously acknowledging in the same response that its search results were inconsistent and that the claim should be independently verified. The model generated a specific, incorrect structural claim rather than declining to characterise an annex it could not retrieve authoritatively. The self-hedging note embedded in the response makes the failure mode transparent but does not prevent the misleading content from being presented.
  • Cited source(s):
Impact for this audience

This finding is diagnostically valuable because the model's response explicitly flags retrieval uncertainty while still generating wrong content — making visible the gap between the uncertainty-signalling subsystem and the generation subsystem. The retrieval layer produced inconsistent signals (different search summaries pointing to different frameworks), the model's own output acknowledged this, and the model generated a specific wrong answer anyway. This points to a calibration gap in the generation layer's use of retrieval confidence: the model should suppress or substantially hedge specific structural claims when its own search results are flagged as inconsistent. This is a targetable post-training intervention: training the model to treat explicit internal retrieval inconsistency as a signal to abstain from specific structural claims.

References — raw findings (per AI model)
This finding also affects
← Previous finding Finding 3. Claude Sonnet 4.6 with web search Next finding → Finding 5. Claude Sonnet 4.6 with web search
Cite this finding

Each finding has a stable Citation ID (RLB-F-… for aggregated case-study findings, RLB-H-… for raw per-model hallucinations) — like a DOI, the ID always resolves to the canonical finding even if URLs change.

Plain text
RegLeg Specialist Panel (2026). "Claude Sonnet 4.6 with web search — AI Labs." Citation ID: . RegLegBrief AI Hallucination Research, published 2026-05-28. https://reglegbrief.com/regulators/j3/sg/mas/notice-637-capital-adequacy-banks-2025/whitepaper/finding/q-NOTICE-637-CAPITAL-ADEQUACY-BANKS-2025-v1-017--sonnet-46-websearch/
APA 7th edition
RegLeg Specialist Panel. (2026). Claude Sonnet 4.6 with web search [Hallucination finding ]. RegLegBrief AI Hallucination Research. https://reglegbrief.com/regulators/j3/sg/mas/notice-637-capital-adequacy-banks-2025/whitepaper/finding/q-NOTICE-637-CAPITAL-ADEQUACY-BANKS-2025-v1-017--sonnet-46-websearch/
Bluebook / OSCOLA (US + UK legal)
RegLeg Specialist Panel, Claude Sonnet 4.6 with web search [], RegLegBrief AI Hallucination Research (May 28, 2026), https://reglegbrief.com/regulators/j3/sg/mas/notice-637-capital-adequacy-banks-2025/whitepaper/finding/q-NOTICE-637-CAPITAL-ADEQUACY-BANKS-2025-v1-017--sonnet-46-websearch/.
BibTeX
@misc{reglegbrief_,
  author    = {RegLeg Specialist Panel},
  title     = {Claude Sonnet 4.6 with web search},
  year      = {2026},
  publisher = {RegLegBrief AI Hallucination Research},
  note      = {Hallucination finding Citation ID: },
  url       = {https://reglegbrief.com/regulators/j3/sg/mas/notice-637-capital-adequacy-banks-2025/whitepaper/finding/q-NOTICE-637-CAPITAL-ADEQUACY-BANKS-2025-v1-017--sonnet-46-websearch/}
}
← Back to case study summary Case study detail →