<- Take me back to my Accountants (CA/PA) (SG) overview
Executive Summary
Accountants in regulatory reporting and financial-statement preparation roles at Singapore-licensed banks and their groups work daily with MAS Notice 637: capital instrument classification, risk-weighted asset computation, capital adequacy disclosures, and the group-level application of the framework. Across the two findings in this cell, an AI model fabricated a parallel holding-company notice and misrepresented MAS's editorial convention for amendment PDFs.
For accountants, both failures convert into a wrong regulatory basis for the reporting work: a fabricated notice in a capital-instrument classification memo misstates the regulatory source for the treatment, and a wrong reading of the amendment's yellow-highlight convention causes accountants to treat editorial annotations as live regulatory text in their accounting policy review.
How AI gets this regulation wrong
Both findings are inference drift. The AI produced confident answers to two questions where the published regulator text resolves the issue: instrument identification (where the model fabricated 'Notice FHC-N637') and amendment-convention reading (where the model imposed a general drafting convention onto a MAS-specific annotation device). Both failure modes are silent in the deliverable; neither output flags that the underlying claim has no regulator-text basis.
| AI's Failure Mode | Count | Affected findings |
|---|---|---|
| Exposed Fabrication | 2 | Finding#1 · Finding#2 |
What that means for your practice
For accountants in regulatory reporting roles on Singapore-licensed banking groups, both findings load onto the same risk: capital adequacy treatments and disclosures recorded against the wrong regulatory basis. The operational consequence is a chain: a wrong regulatory citation in a capital-instrument memo flows into the capital adequacy computation, which flows into the Pillar 3 disclosure, which flows into the audited financial statements and the regulatory return.
| Risk Impact | Count | Affected findings |
|---|---|---|
| Regulatory enforcement / professional liability exposure | 2 | Finding#1 · Finding#2 |
When this affects Accountants (CA/PA)
Accountants in this practice apply Notice 637 across capital instrument classification (Common Equity Tier 1, Additional Tier 1, Tier 2 eligibility memoranda), risk-weighted asset computation under the standardised and internal models approaches, capital adequacy ratio disclosure, and the group-level consolidation of the framework when the banking group operates under a financial holding company. Each of these workstreams produces deliverables that name MAS instruments, cite specific paragraphs, and rely on a correct reading of the amendment-status conventions.
The two findings in this cell map onto AI-assisted research patterns common in this work. First, what regulatory instrument applies to capital adequacy treatment at the financial holding company level. The AI's fabrication of a parallel notice produces a memo-ready citation that does not exist. Second, when reading the 2025 amendment for changes to capital instrument eligibility, classification thresholds, or disclosure requirements, how to interpret yellow-highlighted passages. The AI's wrong characterisation causes accountants to read highlighted annotations as live regulatory changes when MAS's own text says they are annotations that will not appear in the consolidated published Notice.
The findings at a glance
The table below lists each finding from the AI testing on MAS Notice 637 in this cell, showing the topic, the AI's failure mode, and the citation identifier.
Aggregate impact
Considered together, the two findings describe a generation pattern that accountants in Singapore prudential reporting roles should anticipate. The model confidently produces (a) a fabricated parallel regulatory instrument on a question that the source notice resolves through a scope carve-out, and (b) a wrong characterisation of a regulator's editorial convention that maps it onto general drafting practice rather than the regulator's own stated rule.
For prudential reporting integrity, the implications run through the production chain. A capital-instrument classification memo built on a fabricated notice misstates the regulatory basis for the classification. The capital adequacy computation that picks up the misstated basis carries the error into the regulatory return. A Pillar 3 disclosure that references the fabricated notice creates a public-facing statement of regulatory compliance that is sourced to a non-existent instrument. The same compounding occurs with the amendment convention error: capital-instrument memoranda built on the AI's reading treat highlighted annotation text as live regulatory text, producing accounting positions that cite editorial commentary as authority.
What your team should do
Accounting teams in Singapore banking prudential roles should treat AI tools as a search prompt generator on Notice 637 work, with a mandatory verification step against MAS's published text before AI output enters a classification memorandum, capital adequacy computation, disclosure draft, or regulatory return. The findings in this cell show that the failure modes lie in the two question types most attractive to AI-assisted work: instrument identification and amendment-text reading.
Practical safeguards: (a) every MAS instrument citation in a classification memo or disclosure must be matched against the MAS publications portal; AI output is a search prompt only. (b) When reading the 2025 amendment for changes to capital adequacy treatment, pull the reading-convention paragraph (paragraph 3 of the amendment) before logging changes; do not rely on AI-supplied convention descriptions. (c) Maintain an internal catalogue of confirmed MAS instruments applicable to the group, and require classification memos to cite the catalogue rather than an unverified AI-suggested instrument name.
(d) For group-level capital adequacy treatment under a financial holding company structure, anchor the position on the scope paragraph of Notice 637 (paragraph 11.2.2), not on AI-supplied parallel-instrument references.
Where AI tools deliver value in this practice: drafting classification-memo structures, summarising Notice 637 architecture for review against source, and preparing first-pass change logs of amendments for the team to verify against the actual amendment text. The risk concentrates in the moment AI output (instrument numbers, convention statements) is transcribed into prudential reporting deliverables without verification.
How RLB Can Help
RegLeg's published Hallucination Research is available as a free pre-flight check for Singapore banking practitioners working across MAS-supervised entities. Before relying on AI-assisted output for regulatory interpretation, compliance advice, or capital-instrument structuring, practitioners can consult the research to identify where AI tools have demonstrably mis-stated the rules: invented instruments, misread editorial conventions, outdated paragraphs presented as current. The research covers specific MAS instruments and surfaces the exact questions where AI tools have failed, making it a practical reference rather than a general caution.
For firms where multiple teams are working the same regulatory portfolio, RegLeg offers bespoke deep-dives into individual MAS instruments. These engagements go beyond the published findings to examine the full pattern of AI failure modes relevant to the instrument: the question types, the failure mechanisms, and the risk implications for compliance, risk, treasury, legal, and reporting work. The output is designed to be shared across functions and used as a durable reference, reducing duplicated due-diligence effort and creating a consistent internal standard for AI-assisted regulatory work.
RegLeg also develops training and CPD-aligned content for Singapore banking teams. The material translates the failure-mode catalogue into practical guidance on the classes of error practitioners should watch for: confabulated cross-references, version confusion between superseded and current instruments, jurisdiction bleed between superficially similar regimes, and inference-driven elaboration that overstates what an instrument actually requires. Separately, RegLeg offers a confidential review of a firm's existing AI-use policy against the failure-mode catalogue, identifying gaps between the policy's assumptions and the documented evidence of how AI tools perform on Singapore prudential questions in practice.