Practical AI for Australian Small Business

TPB AI Guidance for Tax and BAS Agents

If AI use in client work feels harder to assess than ordinary software, that is understandable. This guide explains the final TPB position, connects it to common practice workflows and shows where professional judgement or advice is still needed.

Last verified: 28 October 2026. References checked against current legislation.

Part of the AI for Accounting Firms and Bookkeeping Practices: A Practical Guide Return to the industry centre →
Editorial Perspective

You run a small practice as a registered tax or BAS agent, and your staff want to use AI for client emails, summaries and research. The pressure is not simply whether the tools save time, but where permission, review and records fit. This guide maps the TPB's published guidance to familiar workflows so you can identify the questions your practice needs to answer. No technical background is needed.

If AI use in client work feels harder to assess than ordinary software, that is understandable. This guide explains the final TPB position, connects it to common practice workflows and shows where professional judgement or advice is still needed.

ⓘ

In short: The Tax Practitioners Board issued TPB(GS) 55/2026 The use of Artificial Intelligence and the Code of Professional Conduct on 22 July 2026. The statement does not ban AI. It says tax practitioners remain responsible for their services and connects AI use to existing expectations concerning competency, reasonable care, confidentiality, records, supervision and quality management.

What the final TPB statement covers

The final statement applies to registered tax agents and BAS agents, which it collectively calls tax practitioners. It provides guidance about using AI while providing tax agent services under the Tax Agent Services Act 2009, the Code of Professional Conduct and the Tax Agent Services (Code of Professional Conduct) Determination 2024.

The TPB also defines the limits of its statement. TPB(GS) 55/2026 says its principles, explanations and examples are not legal advice and do not add rights or obligations beyond those found in the TASA or otherwise at law. It also says the statement is not a comprehensive guide to AI or its capabilities.

The statement recognises that appropriately used AI can improve productivity and efficiency. Its central qualification is that responsibility for the tax agent service remains with the practitioner, including when an AI tool produces part of the work.

That distinction matters for a small practice. The question is not simply whether staff are allowed to open an AI tool. It is how the tool, the information entered, the intended reliance on its output and the practice's review arrangements relate to the TPB guidance.

The questions the TPB puts around an AI task

The TPB lists several general factors that tax practitioners may wish to consider. These include the nature of the task, where information is stored, how the AI uses it, how heavily the output will be relied on and whether the tool is appropriate for the data and purpose.

The statement also points to the tool's capabilities and limitations, the decisions practitioners make while using it and the process for reviewing an output before professional use. These factors make the workflow, rather than the product label, the useful unit of assessment.

For example, an AI feature that rewrites a generic office notice presents different questions from a tool asked to analyse a client's tax position. The second task involves greater intended reliance and may involve client information, technical judgement and independently verifiable claims.

A practice could therefore describe each proposed use by answering four initial questions:

  • What exact activity will the AI perform?
  • What information will staff enter or connect to it?
  • What will the output influence or become?
  • Who will assess the output before it affects client work?

These are organising questions, not a compliance test. The appropriate answers depend on the service, tool configuration, information and circumstances of the practice.

Competency, professional judgement and reasonable care

The competency section of TPB(GS) 55/2026 connects AI use with Code items 7 to 10. As the statement describes them, these items concern competent service, maintaining relevant knowledge and skills, reasonable care when ascertaining a client's state of affairs and reasonable care in applying taxation laws correctly.

The statement says practitioners should not treat AI output as a substitute for their own analysis of a client's circumstances. It notes that AI models may produce inaccurate information or hallucinate, so it points to verification, review and professional judgement throughout the workflow.

The TPB does not give a single formula for reasonable care. Its statement says the assessment depends on all the circumstances, including the nature and scope of the service, whether the practitioner checks the AI output and whether AI was used for a task needing independent verification, particularly outside the practitioner's expertise.

This means a useful internal discussion can start with the consequence of an error. A polished sentence in a client email and a suggested tax treatment may both look plausible, but they do not carry the same intended reliance or technical consequence.

Worked example: drafting a client email

Suppose a staff member asks an AI tool to turn approved notes into a plain-English email explaining the next steps in a BAS matter. Before AI, the staff member would draft the message and a reviewer would compare it with the file. With AI, the practice might choose to preserve that comparison and add a check that the tool has not introduced new claims, dates or instructions.

That review option reflects the TPB's focus on assessing AI output before professional reliance. It does not mean the statement prescribes a particular email approval process. The practice still has to judge what review fits the task and its circumstances.

Worked example: suggesting a tax or BAS treatment

A prompt asking for possible treatment of a transaction goes beyond rewriting approved material. Because the output could influence technical advice, the TPB's discussion of professional judgement, accuracy, reasonable care and independent verification becomes directly relevant.

A practice could treat the output as a research lead rather than a conclusion, then have an appropriately skilled practitioner examine the facts and primary tax sources. That is an example of a firm-level control, not a claim that the TPB has approved a particular workflow.

Records, supervision and quality management

The same TPB AI guidance statement maps AI use to sections 30, 35 and 40 of the 2024 Determination. The statement describes section 30 as concerning records of tax agent services, section 35 as concerning the knowledge, skills and appropriate supervision of entities providing services on a practitioner's behalf, and section 40 as concerning a documented and enforced system of quality management.

For more detail, the AI statement directs readers to TPB(GS) 53/2024 Supervision, competency and quality management under the Tax Agent Services Act 2009. The TPB also maintains a supervision, competency and quality management overview.

The AI statement says verification and review steps should be documented, and says this will assist practitioners to meet their obligations under sections 30 and 40. It does not prescribe an AI register, an approved-use list or a particular escalation matrix.

A small practice may still choose such controls if they help apply its existing supervision and quality arrangements. For example, an internal checklist could record the task, tool, source material, reviewer and outcome for higher-consequence uses without claiming that the TPB mandates that exact format.

Worked example: preparing an internal checklist

An office manager might convert the practice's existing review questions into a short checklist for staff using AI. Before the checklist, staff may make inconsistent decisions about what can be entered and who reviews the result. Afterward, the same questions are raised each time, while the registered practitioner retains judgement about the particular service.

The checklist could distinguish rewriting, summarising, research and treatment suggestions because the intended reliance differs. It could also prompt staff to stop and ask a supervisor when client information, unfamiliar technical issues or uncertain tool settings are involved.

Confidentiality and client permission

The confidentiality section of TPB(GS) 55/2026 points to Code item 6. As the TPB describes it, client information cannot be disclosed to a third party without the client's permission unless there is a legal duty to disclose it.

The statement says entering client information into an AI tool can amount to disclosure to a third party, depending on the tool's configuration and use. When seeking permission, the TPB recommends clearly explaining the proposed disclosure, including who will receive it, where disclosure occurs, where the data will be stored and whether AI may be used.

The statement says permission may be recorded through a signed engagement letter, signed consent or another communication such as an appropriate fact find and consent. It also says a general authority for third-party disclosure may be acceptable, but it does not state that every use of AI must be disclosed to every client.

For the TPB's fuller treatment of Code item 6, see TPB(GS) 26/2014 Confidentiality of client information. The practical question is whether the particular workflow divulges client information to another entity, not merely whether software somewhere contains an AI feature.

Worked example: summarising a client document

A staff member may want to upload a long client document and ask an AI tool for a summary. Before doing so, the practice could examine what information the document contains, where the tool sends or stores it, how the provider uses inputs and whether the proposed use involves third-party disclosure under the TPB guidance.

Removing names may not resolve every confidentiality or privacy question if the remaining facts still identify the client. A practice facing uncertainty about its tool, permission or data handling should seek advice suited to its circumstances rather than assume that a generic AI setting settles the issue.

Privacy Act and tax file number considerations

TPB(GS) 55/2026 says practitioners are responsible for due diligence when using commercial, internally developed or modified AI tools. Its examples include considering information security and whether Privacy Act 1988 requirements are met.

The statement also says additional obligations apply when client information includes tax file numbers under the Privacy (Tax File Number) Rule 2015. It tells practitioners to consider whether the Privacy Act provisions apply to them and to seek their own advice where appropriate, with further information available from the Office of the Australian Information Commissioner.

These privacy points sit alongside, rather than replace, the Code discussion. A tool assessment may therefore need separate questions about confidentiality, privacy coverage, TFNs, storage location and provider handling.

A practical way to organise the practice's response

The TPB statement does not provide a mandatory implementation checklist. A small practice can nevertheless use the following sequence to organise its discussion without treating it as a compliance conclusion:

  1. Describe each proposed AI task and the intended use of its output.
  2. Identify the information involved, including whether it contains client information, personal information or TFNs.
  3. Ask the provider where inputs are sent and stored, how they are used and which settings affect that handling.
  4. Identify the competency, review, verification and record questions raised by the particular task.
  5. Consider how the task fits the practice's existing supervision and quality management arrangements.
  6. Record unresolved questions for the relevant professional adviser, association, insurer, software provider or regulator.

This approach keeps low-consequence drafting separate from technical work that could affect a client's position. It also makes uncertainty visible, which is safer than allowing each staff member to make an undocumented assumption about the tool.

Firm-level options might include approved purposes, restricted data categories, named reviewers or an AI-use record. These can be useful governance choices, but TPB(GS) 55/2026 does not prescribe those specific controls.

Where the TPB guidance stops

The TPB statement explains how existing Code and Determination topics connect with AI, but it does not decide how those provisions apply to every tool or practice. It expressly says it is not legal advice and is not a comprehensive guide to AI capabilities.

Tool configurations, client engagements, professional memberships and practice arrangements differ. Questions about a particular disclosure, consent process, Privacy Act position or standard of care may therefore need advice based on the actual facts.

The statement also notes that APES 110 provides technology-related guidance for members of CA ANZ, CPA Australia or the Institute of Public Accountants, while observing that it is not binding on every tax practitioner. Relevant professional association guidance may add another source for members to examine.

Related reading: our OAIC guidance on AI for businesses.

Related reading: AI for Bookkeepers in Australia: Tools, Limits, and Compliance Risks and AI for Tax Preparation in Australia: What a Small Business Owner Needs to Know.

Related reading: our AI governance by region.

How this was researched

This is desk research: we read named vendor documentation, pricing pages and independent sources, not enterprise demos, and we do not run deployments inside live businesses. Pricing is taken from the vendor's published rates, in the currency the vendor publishes them in; we do not convert prices. Compliance notes reference the legislation and regulatory guidance relevant to each article's region. The question we assess every tool against: could a business with no dedicated IT department pick this up and use it on Monday morning.

Does the TPB ban tax and BAS agents from using AI?

No. TPB(GS) 55/2026 recognises potential productivity and efficiency benefits from appropriately used AI. It also says practitioners remain responsible for their tax agent services and should apply professional judgement before relying on AI output.

Does every use of AI have to be disclosed to a client?

The statement does not say every AI use requires disclosure. Its confidentiality discussion concerns divulging client information to a third party, which can include entering information into an AI tool depending on its configuration and use. See the TPB's AI statement and confidentiality guidance.

Can a general authority cover disclosure to an AI provider?

TPB(GS) 55/2026 says a general authority consenting to third-party disclosure may be acceptable. It also recommends clearly informing the client about the proposed disclosure, including the recipient, location, data storage and possible AI use. Whether a particular authority is suitable requires assessment of the actual arrangement.

Does the TPB require an AI register or approved-use list?

The statement does not prescribe an AI register, approved-use list, access-control scheme or escalation matrix. It discusses review, documentation, supervision and quality management, while leaving practices to assess how those subjects apply in their circumstances. The TPB supervision and quality management guidance provides the broader context.

What does the TPB say about checking AI-generated work?

TPB(GS) 55/2026 says AI-generated content should be verified and reviewed for accuracy throughout the workflow. It also says practitioners should use professional judgement rather than substitute an AI output for their own analysis of the client's circumstances.

What if an AI tool receives tax file numbers?

The TPB statement says additional obligations under the Privacy (Tax File Number) Rule 2015 apply where client information involves TFNs. It directs practitioners to consider whether the relevant privacy provisions apply and seek their own advice where appropriate. The OAIC publishes the primary privacy guidance.

The next step

Read the broader professional ethics guide for AI use in accounting to place the TPB statement alongside wider professional judgement, confidentiality and governance questions. Then compare those questions with the actual AI tasks, information and review arrangements in your practice.

Read the Ethics Guide

Continue your Accounting & Bookkeeping journey

Next Using AI in Management Reporting Workflows
Centre Return to the AI for Accounting Firms and Bookkeeping Practices: A Practical Guide