Checking AI output: competence, due care and R320.11.

AI tools produce confident, well-formatted answers, and some of them are wrong: a tax position that misreads a ruling, a summary that drops a schedule, a reference to a determination that does not exist. APES 110 does not ask accountants to avoid AI. It asks them to decide when relying on it is appropriate, and since 1 January 2025 it says so in terms.

General information only. This guide summarises published professional standards and legislation and is not legal or professional advice. Read the primary sources linked throughout, and take advice on your firm’s specific position.

R320.11: using the output of technology

The technology-related revisions APESB issued in June 2024 added a requirement for members in public practice. When a member intends to use the output of technology in a professional activity, R320.11 requires the member to determine whether that use is appropriate for the intended purpose. The application material at 320.11 A1 lists the factors to consider:

What the technology is doing

The nature of the activity performed by the technology, and how far the work will rely on its output.

Whether you can explain it

Whether the member, or an expert the member can access, can understand, use and explain the technology and its appropriateness for the purpose.

Whether it was tested

Whether the technology has been appropriately tested and evaluated for the purpose, and whether its use for that purpose is generally accepted.

Who oversees it

The firm’s oversight of the technology’s design, implementation, operation, monitoring and updating.

Who can use it

The controls over its use, including procedures authorising user access and overseeing that use.

What goes in

Whether the inputs are appropriate, including the data and the decisions people make while using it.

Section 320 sits in Part 3, so R320.11 applies to members in public practice; Section 220 carries the same factors for members in business. For a firm, it makes adopting an AI tool a documented decision about fitness for purpose, made before anyone relies on the output.

Subsection 113: competence and due care

R113.1 requires members to attain and maintain professional knowledge and skills at the level required to deliver competent professional activities, and to act diligently in accordance with applicable technical and professional standards. The application material at 113.1 A3 says maintaining competence requires a continuing awareness and understanding of technology-related developments relevant to the member’s work. Two further requirements matter for AI:

  • Supervision (R113.2). Members must take reasonable steps to ensure those working under their authority have appropriate training and supervision. A graduate using AI to draft a review note is still working under someone’s authority.
  • Limitations (R113.3). Where appropriate, members must make clients aware of the limitations inherent in their professional activities and explain the implications. Where AI materially shapes a deliverable, its limits may be among them.

A verification workflow that holds up

Check every figure against its source

Totals, balances and dates in an AI summary go back to the workpaper or statement they came from before anything relies on them.

Check every authority against the primary source

Rulings, determinations and section references are confirmed in the ATO legal database or the legislation itself, never taken from the AI’s answer alone.

Review at the right level

The reviewer needs the competence to spot what is wrong. AI output reviewed by someone who could not have produced it has not really been reviewed.

Keep the record

Note on file what the tool did and who checked it. If the work is questioned later, that note is the evidence of due care.

Where deployment posture fits

No architecture removes the duty to check. Two things make checking faster: answers that show the source excerpts they drew on, and a tool that works from the firm’s own files instead of the open internet. Every answer from an Ermos unit carries the excerpts behind it. Deployment also bears on several R320.11 factors directly, because an on-premise unit puts oversight, access controls and inputs inside the firm, where the member can see them. Our page on tax research shows how this works in practice.

Common questions

What does R320.11 of APES 110 require?
When a member in public practice intends to use the output of technology in a professional activity, R320.11 requires them to determine whether that use is appropriate for the intended purpose. The application material lists factors including the extent of reliance, whether the member can explain the technology, how it was tested, the firm's oversight and controls, and the appropriateness of the inputs.
Do accountants have to check AI-generated work?
APES 110 requires diligence under Subsection 113, prohibits being associated with information provided recklessly under Subsection 111, and requires a decision under R320.11 about whether relying on technology output is appropriate. In practice that means checking AI output against source documents and primary law before relying on it.
When did the technology revisions to APES 110 take effect?
APESB issued the technology-related revisions in June 2024. The revisions to Parts 1 to 3, which cover the fundamental principles, the conceptual framework, members in business and members in public practice, took effect on 1 January 2025. The Part 4A revisions apply to audits and reviews of financial statements for periods beginning on or after 1 January 2025.

See Dominion answering from your own documents.

A demonstration takes 30 minutes, uses no client data, and comes with no obligation.

Book a demo