The Answer in 60 Seconds

The defining international incident is Mata v Avianca, Inc. (S.D.N.Y., June 2023), in which a New York federal judge sanctioned attorneys Steven Schwartz and Peter LoDuca after they filed a brief containing six fabricated case citations generated by ChatGPT - citations the AI invented complete with reporter volumes, page numbers and quoted holdings. The case crystallised the professional-indemnity exposure of AI hallucinations: where a professional (lawyer, accountant, financial adviser, engineer, doctor, consultant) relies on AI output without verification and the output is wrong, the professional is responsible for the resulting harm. For Singapore SMEs in regulated advisory work, the framework is well-defined: the Financial Advisers Act 2001 and MAS Notice FAA-N16 (reasonable basis for recommendations) for financial advice; the Legal Profession Act 1966 for legal services; the Accountants Act 2004 for accounting practice; the Healthcare Services Act 2020 for healthcare services. The exposure operates through the existing professional negligence framework anchored in Spandeck Engineering v DSTA [2007] SGCA 37. This article sets out where the exposure crystallises, which insurance covers respond, and the verification protocol that contains the risk.

The Sourced Detail

The use of generative AI in professional advisory work in Singapore is now broad: legal research and drafting, accounting and tax analysis, financial-advice ideation, engineering calculation checks, medical literature review, management consulting deliverables. The productivity benefit is real and visible; the failure mode - the AI fabricates a citation, a calculation, a regulatory reference, or a clinical fact - is documented globally and is increasingly documented in Singapore-adjacent contexts.

The professional-indemnity exposure depends not on whether the professional used AI - the use is largely lawful in itself - but on whether the professional verified the AI output to the standard the relevant duty of care requires.

The international anchor: Mata v Avianca

In Mata v Avianca, Inc. (S.D.N.Y., June 2023), attorneys Schwartz and LoDuca filed an opposition brief on behalf of a personal-injury claimant that cited six federal cases purporting to support their argument. None of the six cases existed. ChatGPT had generated them - case names, reporter citations, court holdings, and even fabricated direct quotes from supposed judges. When opposing counsel and the court raised the discrepancy, Schwartz initially asked ChatGPT to confirm the citations; the model produced false confirmations, complete with apparent extracts from the non-existent decisions.

District Judge P. Kevin Castel imposed sanctions on Schwartz, LoDuca and their firm in a written opinion. The principle established: an attorney is responsible for what they file, regardless of the technology that produced the draft. The same principle applies, with appropriate adaptation, across every regulated profession.

The Singapore framework: Spandeck negligence

In Spandeck Engineering (S) Pte Ltd v Defence Science & Technology Agency [2007] SGCA 37, a five-judge Court of Appeal led by Chief Justice Chan Sek Keong established the single two-stage test that governs negligence claims in Singapore: factual foreseeability (the threshold), then a combined analysis of legal proximity and policy considerations. The test applies uniformly to all categories of negligence including professional negligence and pure economic loss.

A claimant harmed by an AI-fabricated citation, calculation, or recommendation in professional work pleads Spandeck against the professional. The proximity analysis is straightforward - the client engaged the professional for the advice; the policy analysis is also straightforward - imposing the duty of verification on professionals is consistent with the regulatory structure each profession sits within.

The regulated-activity framework

Each Singapore advisory profession has its own statutory and regulatory architecture, within which the AI-output exposure crystallises.

Financial advice. The Financial Advisers Act 2001 governs the conduct of FAs. MAS Notice FAA-N16 requires a reasonable basis for any recommendation on investment products - the FA must have considered the client's investment objectives, financial situation and particular needs, and have a reasonable basis for the recommendation given those factors. An AI-generated recommendation that was not verified against the client's actual circumstances breaches the Notice; the responsibility is the FA's, not the AI's. MAS Notice FAA-N03 sets the information-to-clients and product-disclosure standard.

Legal services. The Legal Profession Act 1966 regulates Singapore lawyers; section 75A and the Legal Profession (Professional Indemnity Insurance) Rules at LPA1966-R11 make professional indemnity insurance mandatory for solicitors in practice. The compulsory minimum is S$1 million per claim for a sole proprietor or partnership and S$2 million per claim for a Law Corporation or LLP, with the scheme operative since 1991. A lawyer who files an AI-hallucinated citation has failed in the duty to the court and to the client; the PII responds to the resulting claim subject to the policy wording and the compulsory scheme terms.

Accounting practice. The Accountants Act 2004 regulates public accountants. Section 28 requires accounting firms to maintain professional indemnity insurance to a formula: the higher of S$1 million per claim, or S$500,000 per corporate practitioner, or 2.5 times the firm's gross income (capped at S$50 million). An AI-fabricated tax position or audit finding is an accountant's responsibility under the verification standard the profession applies.

Engineering. The Professional Engineers Act 1991 regulates Singapore PEs; section 34 imposes liability-insurance requirements. An AI calculation error that flows through into a structural design is the PE's responsibility under the verification standard.

Architecture. The Architects Act 1991 regulates Singapore architects; section 24 imposes liability-insurance requirements. Similar AI-verification analysis applies.

Healthcare services. The Healthcare Services Act 2020 regulates healthcare service providers; the Medical Registration Act regime regulates registered medical practitioners. An AI-generated clinical decision-support output that is not verified by the clinician is the clinician's responsibility under professional duty standards.

Where the hallucination exposure crystallises

Five recurring scenarios in Singapore advisory work.

Scenario 1: Fabricated legal citations. The Mata pattern. A Singapore lawyer using a generative AI tool for research receives a brief or memorandum containing case citations that do not exist or that misstate the holdings. The Legal Profession Act and the Rules of Court duties apply; the PII responds to any resulting professional negligence claim.

Scenario 2: Fabricated regulatory references. A compliance professional, accountant, or financial adviser receives an AI-generated reference to a regulation, notice or section that does not exist or that has been mis-cited. The misstatement flows into advice given to the client.

Scenario 3: Mis-stated facts about a real person, company, or transaction. An advisor uses AI to summarise a public-record fact set; the AI fabricates a fact that, if relied upon, harms a third party (or harms the client). Defamation, negligence, or breach of fiduciary duty may follow.

Scenario 4: Computation or calculation errors. An engineer, accountant, or actuary uses AI to perform or check a calculation; the AI generates a confident but incorrect answer. The downstream use of the calculation - in a design, an audit opinion, or a recommendation - carries the consequences.

Scenario 5: Mis-attribution of a published source. An advisor cites a published source (a regulator's guidance, an industry standard) on the strength of an AI summary; the AI has paraphrased or invented content that the source does not contain. The client relies on the misattributed content.

The verification protocol that contains the risk

Four-step verification, consistently applied, removes most of the hallucination exposure.

  1. Treat AI output as draft, not final. The same standard applied to junior-staff work product.
  2. Independently verify every primary-source citation. Statutes, cases, regulations, notices, official guidance - check each against the actual source. AI confidence is not corroboration.
  3. Re-perform numerical calculations. Either by hand or in an independent tool. AI calculation errors are not random; they cluster on particular question types.
  4. Disclose AI use to the client where the engagement terms require it. Many professional engagement letters are being updated to address AI use; the disclosure norm is rising.

The protocol is the same in spirit as the historical professional standard for verifying junior-staff work product before it leaves the firm under the supervising professional's name. The AI does not change the standard; it changes the source of the draft.

Insurance covers that respond

Professional indemnity (PI). PI is the principal cover. PI is claims-made and responds to third-party claims arising from professional services. Two wording checks apply:

  • Whether the policy carries an AI-related exclusion (some 2024-2025 wordings introduced these; many did not).
  • Whether the policy's "professional services" definition extends to AI-assisted work product. Most wordings do, but the question should be confirmed.

Cyber liability. Cyber typically does not respond to a hallucination-driven professional negligence claim, because the loss is not a cyber event in the policy's sense. Where the hallucination involves a confidential-information exposure (the AI fabricates content about a client's confidential matter), cyber may respond as a privacy event.

Errors and omissions / Technology E&O. For technology professionals, Tech E&O is the primary cover and addresses both technology-product defects and AI-related output errors.

Common Mistakes / What Goes Wrong

  1. Trusting AI output as if it were verified. The Mata pattern.
  2. Re-prompting the same AI for citation confirmation. Hallucinations compound.
  3. No engagement-letter language addressing AI use. Client may have non-disclosure expectations.
  4. PI cover with AI exclusion that has not been read. The 2024-2025 wording change matters.
  5. No documented verification protocol in the firm's standards.
  6. Different practitioners applying different standards. The firm's protocol is the protocol.
  7. AI used for the deliverable to the client without supervision. The supervisor is responsible.
  8. No PDPA assessment when AI tools touch client personal data.
  9. No CPD on AI verification practice. Professional bodies' guidance is evolving.
  10. No post-incident review of the verification failure.

What This Means for Your Business

  1. Adopt a verification protocol - treat AI output as draft.
  2. Independently verify every primary-source citation.
  3. Re-perform calculations by hand or in an independent tool.
  4. Confirm your PI wording does not exclude AI-assisted work.
  5. Update engagement letters to address AI use disclosure.
  6. Train staff on the verification protocol annually.
  7. Document the protocol in the firm's standards manual.
  8. Track post-incident learning - a hallucination caught is a control success; a hallucination missed is a control failure.

Questions to Ask Your Adviser

  1. Does our PI wording contain AI-related exclusions, and if so, what is the scope?
  2. Is our "professional services" definition broad enough to cover AI-assisted work?
  3. For any AI-related professional negligence claim, what is your support model at notification?
  4. Have you seen any Singapore-market AI-related PI claims, and what were the patterns?
  5. Are affirmative AI endorsements available for our PI cover at our next renewal?

Related Information

Published 22 May 2026. Source verified 22 May 2026. COVA is an introducer under MAS Notice FAA-N02. We do not recommend insurance products. We provide factual information sourced from primary regulators and route you to a licensed IFA who can match a policy to your specific situation.