Part of the AI Regulation News hub.
The Tax Practitioners Board has explained how existing Code of Professional Conduct obligations apply when a registered tax agent or BAS agent uses AI
Australia did not write an AI rule for tax agents. It wrote a map showing which statutory Code items already reach AI use, and the busiest one on the map is the quality-management obligation.
Bottom line: A Guidance Statement, not legislation. It creates no new obligation. The TPB says in its own disclaimer that the principles, explanations and examples in it do not constitute legal advice and do not create additional rights or legal obligations beyond those contained in the TASA or which may exist at law. What binds is the Tax Agent Services Act 2009 Code and the 2024 Determination it points to.
Who this affects: Registered tax agents and BAS agents in Australia, the partners and supervising agents responsible for their systems of quality management, and US CPAs and accounting-firm risk leads who want a working model of how a tax regulator frames AI supervision.
Issue date: 22 July 2026, on both the Guidance Statement itself and the TPB media release. It followed an exposure draft released 24 March 2026 with submissions closing 21 April 2026. We are covering it in August because it has not been covered here, not because it is new.
What changed: Nothing in the statute. What is new is a published TPB view mapping AI use onto named provisions: Code items 6 through 10, and sections 30, 35 and 40 of the Tax Agent Services (Code of Professional Conduct) Determination 2024.
Analysis: The centre of gravity is not the AI. It is section 40 of the Determination, the documented system of quality management. The TPB routes AI review, verification and contestation steps into that existing system rather than standing up a parallel AI control. A firm with a weak quality-management file does not have an AI problem; it has a section 40 problem that AI makes visible.
Primary sources: TPB(GS) 55/2026 (full text) · TPB media release, 22 July 2026 · PDF of the Guidance Statement
- Instrument (EN)
- TPB(GS) 55/2026 The use of Artificial Intelligence and the Code of Professional Conduct
- Authority
- Tax Practitioners Board (TPB), Australia
- Jurisdiction
- Australia. Applies to entities registered with the TPB under the Tax Agent Services Act 2009.
- Status
- Final. Published 22 July 2026 after an exposure draft of 24 March 2026 and consultation closing 21 April 2026.
- Bindingness
- Not binding of itself. It is a Guidance Statement setting out the Board view. The TPB states it does not create additional rights or legal obligations beyond the TASA. The Code items and Determination sections it describes are the binding law.
- Provisions engaged
- Code items 6, 7, 8, 9 and 10 (section 30-10 TASA); sections 30, 35 and 40 of the Tax Agent Services (Code of Professional Conduct) Determination 2024.
- Issue date / next deadline
- Issued 22 July 2026. No compliance deadline is set in the document. The TPB announced a supporting webinar, AI and the Code: What every tax practitioner needs to know, for 25 August at AEST 12 pm.
- Related regime referenced
- Privacy Act 1988 (Cth), the Australian Privacy Principles, and the Privacy (Tax File Number) Rule 2015.
- Primary source
- https://www.tpb.gov.au/tpbgs-552026-use-artificial-intelligence-and-code-professional-conduct
Why a July document is showing up here in August
This desk covers it now because it was not covered before. Nothing about the instrument changed between 22 July and today. It was issued on 22 July 2026 and the last-modified stamp on the TPB page is the same date, so there is no quiet revision to report.
That matters for how you read the rest of this piece. Trade coverage of the Australian tax-agent regime tends to date guidance from the commentary rather than the instrument. Both the Guidance Statement page and the TPB media release carry 22 July 2026. That is the date to use.
One live consequence: the TPB scheduled a supporting webinar for 25 August at AEST 12 pm. If you read this the day it posts, that session has not happened yet.
It is guidance about old duties, not a new AI duty
The distinction carries the whole analysis, so it is worth being exact about it. TPB(GS) 55/2026 does not impose anything. Its own disclaimer says the principles, explanations and examples in it do not constitute legal advice and do not create additional rights or legal obligations beyond those contained in the TASA or which may exist at law, and it directs the reader to the TASA for the precise content of the legislative requirements. It also says it does not exhaust, prescribe or limit the scope of the TPB powers in the TASA.
So what is the document for? It is the Board setting out its view of how provisions that already exist reach a practice that is new. Paragraph 1 says the purpose is merely to provide guidance to tax practitioners in relation to their statutory obligations under the TASA when using AI in the provision of tax agent services. A registered agent is expected to have regard to it. A registered agent is bound by the Code items behind it.
The practical test of that difference: if the TPB ever acts against an agent over AI use, the charge will read as a Code item or a Determination section, not as a breach of TPB(GS) 55/2026. Our reading, not a statement in the document.
The provisions the document actually names
The Guidance Statement carries two provision tables. Table 1, headed competency, reproduces Code item 7 (a tax agent or BAS service you provide, or that is provided on your behalf, is provided competently), Code item 8 (maintain knowledge and skills relevant to the services provided), Code item 9 (reasonable care in ascertaining the state of a client affairs), and Code item 10 (reasonable care to ensure the taxation laws are applied correctly). It then reproduces three sections of the 2024 Determination: section 30 on keeping records that correctly record the tax agent services provided, section 35 on supervision of entities providing services on your behalf, and section 40 on the system of quality management.
Table 2 is a single row: Code item 6, confidentiality. Unless there is a legal duty to disclose, information relating to a client affairs is not to go to a third party without the client permission.
Beyond those, paragraph 28 notes there are further obligations under the Code and Determination touching honesty and integrity, independence and other responsibilities, but it does not tabulate them or map them to specific AI facts. Treat that paragraph as a pointer, not a list.
What an agent has to be able to show
Competence in the tool. The Guidance Statement says tax practitioners remain accountable for the accuracy of information and advice given to clients, that it is important to understand the capabilities and limitations of AI tools, and that AI output should be assessed and supplemented by professional judgement before being relied on. It is blunt about why: AI models may hallucinate or provide inaccurate information and cannot be relied on as a replacement for tax knowledge, experience or expertise.
Documentation of the AI step. Paragraph 16 asks practitioners to verify and review AI-generated content for accuracy throughout each step of the workflow and to establish processes to understand and contest AI decisions and outputs, and then says each of these steps should be documented. It ties that directly to sections 30 and 40 of the Determination. This is the sentence most likely to be read back to a firm later, because it converts a habit into a record.
Confidentiality before the prompt, not after. Paragraph 23 states that in the context of AI, permission has to be obtained from each client before divulging client information to a third party, and it says entering client information into AI models and tools can be such a disclosure depending on how the tools are configured and used. The recommended practice is to tell the client to whom and where the disclosure will be made, where data will be stored, and whether AI tools may be used. Acceptable permission includes a signed letter of engagement, signed consent, or a fact find and consent, and the document says a general authority consenting to disclosure to third parties may also be acceptable.
Supervision and review inside a documented system. Section 35 of the Determination reaches entities providing tax agent services on your behalf and expressly refers back to your system of quality management. Section 40 is the quality-management obligation itself. The two are quoted in Table 1 rather than paraphrased, which is a signal about where the Board expects the evidence to sit.
One caution on scope. Section 35 is written about entities providing services on your behalf. The document does not say an AI tool is such an entity, and neither do we. What it does is place AI review steps inside the same quality-management system that supervision is judged against.
The read-across for a US CPA
Set aside the Australian statute for a moment and look at the shape. A tax regulator was asked what the rules are for AI, and answered by naming competence, reasonable care, confidentiality, record-keeping, supervision and documented quality management. It did not create an AI licence, an AI register, a disclosure duty to the revenue authority, or an approved-tools list. That is the clearest statement any tax regulator has made that AI use is assessed against existing statutory duties rather than a separate AI rule.
Circular 230 already contains the equivalent furniture: competence, diligence as to accuracy, reliance on others, and the procedures-to-ensure-compliance obligation that sits on firms rather than individuals. Our circular-230-ai-tax-practitioner-checklist entry works through that inventory. If you want a forecast of the questions an OPR examiner would ask about AI, the TPB tables are a reasonable proxy for the categories.
We are not predicting IRS behaviour. The IRS and OPR have said nothing here, TPB(GS) 55/2026 has no application to US practice, and Australian guidance does not interpret Circular 230. The transferable item is the analytical move, not the authority.
There is a second-order point worth flagging for firms operating in both places. If your Australian and US arms answer the same AI-governance question with different documents, the Australian answer is now the one with a published regulator view attached to it.
What we did not verify
Opened and read in full: the TPB(GS) 55/2026 web version at tpb.gov.au, all 32 numbered paragraphs, both provision tables, the disclaimer, the document history, the further-information table and all 19 footnotes. Also opened: the TPB media release of 22 July 2026, which supplied the issue date, the Chair quotation and the 25 August webinar detail. Every quotation in this article was character-matched against that extracted text.
Not opened: the linked PDF of the Guidance Statement, so we cannot confirm the web and PDF versions are identical. Not opened: the TASA itself, the Tax Agent Services (Code of Professional Conduct) Determination 2024, TPB(GS) 53/2024 on supervision and quality management, TPB(GS) 52/2024 on client records, TPB(GS) 26/2014 on confidentiality, the APESB technical alert, the Privacy Act 1988 or the Privacy (Tax File Number) Rule 2015. Every provision text quoted here is quoted as the Guidance Statement reproduces it in its tables, not as verified against the legislative instrument. Not opened: the submissions made to the March 2026 exposure draft, so we cannot say what changed between draft and final.
We refuse to claim: that this instrument creates any obligation, that it is legislation, or that it applies outside Australia. We refuse to state any sanction, penalty or enforcement consequence for AI use, because the document states none; it links to separate TPB guidance on failing to comply with your obligations, which we did not open. We refuse to say the TPB has taken or will take action against any agent over AI. We refuse to say what the IRS, Treasury or the Office of Professional Responsibility will do with Circular 230. And we make no claim about how many Australian registered agents use AI, because the document supplies no such figure and we have no denominator.
Read this as a supervision document, not an AI document. The TPB view is that AI work product has to survive the same competence, confidentiality, record-keeping and quality-management provisions that already govern everything else a registered agent does, and that the review and contestation steps around AI output should be documented inside the existing system of quality management. A US firm can run the same exercise against Circular 230 today without waiting for the IRS to publish anything. If your AI review steps exist only as practice and not as a file, the Australian framing says you have nothing to produce.
Source File
https://www.tpb.gov.au/tpbgs-552026-use-artificial-intelligence-and-code-professional-conduct
Open the TPB(GS) 55/2026 page at tpb.gov.au and confirm three things. First, the header reads Issued 22 July 2026 and Last modified 22 July 2026, and the Document history paragraph records the 24 March 2026 exposure draft and the 21 April 2026 close of submissions. Second, the disclaimer states the document does not create additional rights or legal obligations beyond those contained in the TASA. Third, Table 1 lists Code items 7, 8, 9 and 10 plus sections 30, 35 and 40 of the Determination, and Table 2 lists only Code item 6. Then open the TPB media release for the same date to confirm the issue date independently.
You must establish and maintain a system of quality management, in relation to the provision of tax agent services by you, or on your behalf, which is designed to provide you with reasonable confidence that you are complying with the Code. You must document and enforce the policies and procedures of your system of quality management. ยท Section 40 of the Determination, as reproduced in Table 1 of TPB(GS) 55/2026, 22 July 2026
FAQ
Does TPB(GS) 55/2026 create new obligations for tax agents using AI?
No. It is a Guidance Statement. Its disclaimer says the principles, explanations and examples in it do not create additional rights or legal obligations beyond those contained in the TASA or which may exist at law, and it refers the reader to the TASA for the precise content of the legislative requirements. The obligations that bind are the Code items in section 30-10 of the Tax Agent Services Act 2009 and the sections of the Tax Agent Services (Code of Professional Conduct) Determination 2024 that the document reproduces.
Which Code items does the document actually engage?
Table 1 reproduces Code items 7, 8, 9 and 10 alongside sections 30, 35 and 40 of the 2024 Determination, under the heading of competency. Table 2 reproduces Code item 6, confidentiality. Paragraph 28 notes that other Code and Determination obligations exist, including honesty and integrity and independence, but does not tabulate or map them.
Does putting client information into an AI tool count as disclosure to a third party?
The Guidance Statement says it can. Paragraph 23 states that permission has to be obtained from each client before divulging client information to a third party, which it says can include entering client information into AI models and tools, depending on how those tools are configured and used. Paragraph 21 defines a third party for TASA purposes as any entity other than the client and the tax practitioner. The document recommends telling the client to whom and where disclosure will be made, where data will be stored, and whether AI tools may be used.
Does any of this apply to a US CPA?
No. It applies to entities registered with the TPB under Australian law and it says nothing about US practice. The interest for a US reader is structural: a tax regulator answered the AI question entirely through existing competence, care, confidentiality, record-keeping and quality-management duties rather than a standalone AI rule. Whether the IRS or the Office of Professional Responsibility takes the same approach under Circular 230 is not something this document decides, and we do not assert it.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.