What the TPB's AI guidance means for Australian tax practitioners
The Tax Practitioners Board has issued guidance on how registered tax practitioners should use artificial intelligence while continuing to meet their obligations under the Code of Professional Conduct. This page explains, in practical terms, what that means for an Australian accounting firm adopting AI in its tax work.
The central point is straightforward: using AI does not transfer a practitioner's responsibilities to the software. The obligations follow the practitioner, not the tool.
Published 11 August 2026. Last reviewed 11 August 2026. Technically reviewed by the Taxpartna tax team.
Artificial intelligence is now part of many Australian tax practices, whether through research assistants, document processing, drafting tools or quality assurance review. The Tax Practitioners Board has responded with guidance on how AI interacts with a registered practitioner's obligations.
This page summarises the practical considerations for firms. It is general information, not a substitute for reading the TPB guidance and the Code of Professional Conduct directly, and it is not legal advice.
AI use does not transfer responsibility to software
The starting position is the most important one. Under the Tax Agent Services Act 2009 and the Code of Professional Conduct, a registered tax practitioner is responsible for taking reasonable care to ascertain a client's state of affairs and to apply the tax laws correctly.
Software does not hold a registration, cannot be a party to the engagement, and cannot be accountable to the client or the Board. When a firm uses AI, the responsibility for the work remains with the practitioner. AI changes how the work is done, not who is answerable for it.
Understanding the capabilities and limitations of AI
A practitioner using AI should understand, at a working level, what the tool can and cannot do.
- AI is well suited to reading varied documents and recognising that differently labelled items refer to the same thing.
- AI language models are probabilistic, so the same question can produce different answers, and a confident answer is not necessarily a correct one.
- AI trained mainly on overseas content can be unreliable on Australian concepts such as Division 7A, Section 100A, trust distributions, franking accounts and the base rate entity test.
- AI cannot know what is not in the file. If a relevant fact was never documented, no tool will find it.
Reviewing AI-generated outputs
AI output should be treated as an input to the practitioner's work, not a conclusion. A reasonable process is to check that the output is supported by the source material, that any citations are accurate and current, and that the reasoning is sound before relying on it.
Tools that show their working, including the values compared and the documents they came from, make this review far more efficient than tools that present an answer without evidence.
Applying professional judgement
AI can surface a matter, but deciding what the matter means for a particular client, with their particular circumstances and history, is professional work. Professional judgement, and the final conclusion, remain with the practitioner. AI does not independently reach the professional conclusion.
Confidentiality and client information
Confidentiality is one of the clearest areas of risk. Before adopting an AI tool, a firm should understand:
- whether client data is sent to a third-party model provider
- whether data is retained, and for how long
- whether data is used to train models
- where data is physically processed and stored
- how the arrangement sits against the firm's confidentiality and Privacy Act obligations
A self-hosted or private model deployment keeps data within a controlled environment. A thin wrapper over a public model API generally does not. This distinction matters for a firm holding TFNs and client financial records.
Supervision and control
Where staff use AI tools, the firm should have arrangements for supervision and control that are appropriate to the risk. That includes knowing which tools are in use, how they are used, and who is responsible for reviewing the output before it is relied upon or sent to a client.
Quality management
AI use should sit within the firm's quality management arrangements rather than outside them. Consistent processes for how AI is used, reviewed and documented help ensure the technology supports quality rather than introducing new and unmanaged risk. Our page on the TPB quality management system requirements for tax agents covers this area in more detail.
Record keeping
Where AI contributes to the work, a firm should be able to show what was done and how the output was reviewed. A defensible file note records the checks performed, the matters identified, and how the practitioner responded, so the review can be understood later without reconstructing the whole job.
Where deterministic logic differs from AI interpretation
A practitioner assessing an AI tool should ask which parts of the output are generated by a language model and which are produced by deterministic computer logic.
Deterministic logic applies a defined rule to defined inputs and returns the same result every time, which suits exact comparisons, reconciliations and tolerance testing. AI interpretation suits reading and connecting documents. Tools that use each capability for what it is suited to tend to be more reliable, and easier to review, than tools relying on AI alone.
Questions firms should ask before adopting AI tax software
- Where does client data go, and is it retained or used for training?
- Does the tool show its working with source references?
- What does the tool do when information is missing or unclear?
- Which parts of the output are deterministic and which are AI-generated?
- Can the output be audited and retained in the client file?
- Is the tax content genuinely Australian?
- Does the tool keep the practitioner in control of the final conclusion?
How Taxpartna approaches human-reviewed tax quality assurance
Taxpartna is designed to support this position rather than work around it. It augments the professional review by locating, comparing and organising relevant information in the file for the practitioner to assess.
It combines AI-assisted document analysis, Australian tax expertise and deterministic computer logic for defined tests and comparisons. Client data is processed on a self-hosted private model, nothing is shared with third-party model providers, and every flagged item is presented for the practitioner's assessment.
Taxpartna does not provide tax advice, does not lodge returns, and does not sign off. The registered tax practitioner retains responsibility for calculations, professional assessment, conclusions and final sign-off. You can read more about our approach to AI tax software in Australia and how it fits into tax quality assurance sign-off.
Read the primary sources
- TPB guidance on the use of artificial intelligence and the Code of Professional Conduct
- TPB Code of Professional Conduct
Explore Taxpartna
- AI tax software in Australia
- Tax quality assurance sign-off
- Taxpartna security
- Taxpartna review features
Frequently asked questions
The TPB guidance does not prohibit the use of artificial intelligence. It sets out how existing obligations under the Tax Agent Services Act 2009 and the Code of Professional Conduct continue to apply when AI is used, including responsibility, competence, confidentiality and supervision.
This page provides general information about the use of artificial intelligence by Australian tax practitioners and about the Taxpartna platform. It does not constitute tax, legal or professional advice, and it is not a substitute for the TPB guidance or the Code of Professional Conduct. Taxpartna is a quality assurance assistance tool designed for use by registered tax practitioners. Taxpartna does not provide tax advice, tax agent services or BAS agent services. All professional judgements and sign-off decisions remain the responsibility of the registered tax practitioner.