The Auditor Owns the Outcome: Singapore's Accounting Regulator on Using AI Responsibly, October 2026

On October 2, 2026, Singapore's Accounting and Corporate Regulatory Authority, ACRA, announced Audit Practice Guidance No. 1 of 2026, "Using Artificial Intelligence Responsibly." ACRA describes itself as the regulator of business registration, financial reporting, public accountants and corporate service providers. The guidance sets out considerations for auditors and audit firms on using AI in their work, and on evaluating information that AI prepared or helped prepare. It says those considerations are not exhaustive. Parts of it follow, not the whole.

In Singapore, the guidance says, the adoption of AI tools among what it calls accounting entities increased from 6% in 2023 to 11% in 2024, as noted in the Accounting Entities Survey 2025. Auditors are likely to meet AI in two ways, it says: using it to support audit procedures, and auditing information that AI prepared or assisted. Used appropriately, AI can improve efficiency and the quality of audit work, it says, but it also introduces risks that auditors and audit firms need to understand and manage. That, in its words, creates "a new responsibility for the profession," and "the responsible use of AI must become as important as the effective use of AI."

Integrity, and the Same Rigor

Using AI does not diminish an auditor's responsibility to act with honesty and integrity, the guidance says. Auditors should not use AI to misrepresent work performed, should not use it to fabricate or manipulate information, and should not rely selectively on its outputs "to support predetermined conclusions." Where AI has been used, they should accurately represent the work performed and the basis for their conclusions.

AI does not replace professional judgment, the guidance says, and auditors must apply the same rigor to its outputs as to any other source of audit evidence or information. They should exercise professional skepticism, the guidance says, because although AI can process vast amounts of information quickly, "it can hallucinate, reinforce bias and misinterpret context, sometimes in ways that are not immediately obvious." They should be mindful of automation bias too.

It sets two kinds of work apart. Where AI follows predefined steps, such as extracting information from documents, auditors should consider whether its outputs are complete and accurate. Where the work involves professional judgment, such as risk assessment or generating audit documentation, they "should, where appropriate, verify AI-generated outputs against underlying sources and assess the reasonableness of the resulting analysis or conclusion."

Its questions for auditors include one about a junior colleague. "If the same analysis had been prepared by AI instead of a junior member of your team, would you have reviewed it with the same rigour and professional scepticism?"

The Auditor Owns the Outcome

"AI can assist, but the auditor owns the outcome." Auditors remain accountable for the work performed, the conclusions reached and the audit opinion expressed, the guidance says, even where AI has supported parts of the engagement. "This responsibility cannot be delegated to AI."

It sets this out as "Consistent with the approach taken in" SSA 620, on using the work of an auditor's expert, which it says makes clear that the auditor's responsibility for the audit opinion is not reduced by using an expert's work. "If the auditor remains fully responsible when using the work of an auditor's expert, the same principle applies when using AI." AI, it adds, is only a tool, and cannot exercise professional judgment, apply professional skepticism or assume responsibility for the audit.

The guidance points to recent incidents involving international accounting firms, which, it says, highlighted the risk that AI-generated outputs "can contain fabricated references, fictitious case studies and other inaccuracies that may not be immediately apparent." The firms withdrew or corrected the reports, it says, and in one case agreed to partially refund the client. The incidents illustrate, in its words, that professionals and organizations "remain responsible for reviewing, approving and issuing the reports, even where AI has been used to assist in the preparation."

Transparent in How They Use AI

The use of AI, the guidance says, should not obscure how audit work was performed or how conclusions were reached. Audit documentation should show how AI-assisted outputs were reviewed, how professional judgment was applied, and how the conclusions were supported by sufficient appropriate audit evidence. Where appropriate, it "may also include the prompts used, key assumptions adopted, significant revisions made following human review, and the rationale for accepting or rejecting AI-generated recommendations." Such documentation, it says, lets engagement reviewers and regulators understand how professional judgment was exercised. Audit firms should set clear policies for the documentation, review and oversight of AI-assisted audit work.

Transparency, it says, may also extend beyond the engagement team. Where appropriate, auditors should be able to explain to those charged with governance how AI was used, the safeguards applied and how judgment was exercised. Among its questions is this one. "Where AI-generated outputs could not be fully verified, have you clearly documented the limitations and reflected them appropriately in your conclusions?"

Confidentiality, and the Same Standard

Auditors should understand how an AI platform processes, stores, retains and shares data, the guidance says, including whether what they enter may be retained or used to train the underlying model. "Where there is uncertainty, auditors should avoid entering confidential or client-identifiable information." They should use only AI tools approved under their firm's requirements, and firms should set policies on which tools may be used and what information may go into them.

In its conclusion, the guidance turns to the profession itself. "Auditors have long expected businesses to explain and support how information was prepared and how the underlying figures were derived. The profession should apply the same rigour to its own use of AI."

That turns the auditor's own question around. Could the file of an audit done with AI explain how its figures were derived as clearly as an auditor asks a client to?