Guide
AI for accounting firms in Australia: the tools and the rules
By Brett Raven, APAC Lead & Board Member, AIREUpdated
Yes, an Australian accounting firm can use AI on client work, and it is already arriving inside Xero, MYOB and practice management software. Two sets of rules decide how. The Tax Practitioners Board's July 2026 guidance says a registered agent needs the client's permission before client information goes to a third party, which can include an AI tool, and should check and document what the tool produces. The APESB, which sets the ethics code for members of CA ANZ, CPA Australia and the IPA, says members should disclose when they use AI tools and supervise that use. So the firm needs a business plan that doesn't train on its data, an engagement letter that covers AI, a written rule on what staff may paste in, and a person who checks every figure before it leaves the firm.
Where AI helps in an accounting firm
The value sits in the reading and writing around the numbers. Much of a junior's week goes on reading source documents and drafting the same letters again. That work suits AI well, as long as someone who could have done it checks it.
Work it does well
- First drafts of client letters and advice. The covering note for a return, or a first cut of an advice letter from the partner's notes. The partner edits instead of starting from a blank page.
- Reconciliation and variance commentary. Give it the reconciliation or two periods of figures and ask for the plain-English account of what moved. Someone still ties each sentence back to the ledger.
- Research summaries. A first read of a ruling or a long piece of legislation, so the accountant knows where to look. The source is what gets cited, never the summary.
- Meeting notes and file notes. A client meeting turned into a file note and a task list, reviewed by the person who was in the room.
- Inherited spreadsheets. What a workbook does and where its assumptions sit, explained before anyone relies on it.
Work it shouldn't do
It shouldn't be the source of a figure that goes to a client or the ATO. General assistants write text that reads as right, and that includes a total that doesn't add up. The APESB's technical alert warns that AI "at times can hallucinate and present convincing answers that are not accurate." Keep the calculations in the ledger and the tax software. Use AI to find, summarise and draft.
The tools: general assistants and the AI in your software
A firm ends up with two kinds of AI. General assistants sit beside the work and take whatever you paste in. Platform AI is built into software that already holds client data, such as Xero, MYOB or your practice management system. You'll usually want one of each.
| Tool | What it does in a firm | What the vendor says about your data |
|---|---|---|
| ChatGPT Business (OpenAI) | Drafting, summarising and custom GPTs built on the firm's own templates. A separate workspace, outside your ledger and practice software. | OpenAI doesn't train on business data by default, unless you opt in, for example through feedback. Workspace admins can view, export and delete members' conversations. |
| Microsoft Copilot (formerly Microsoft 365 Copilot) | Works inside Word, Excel, Outlook and Teams, on the files and email your Microsoft 365 tenant already holds. | Prompts, responses and the data it reads through Microsoft Graph aren't used to train foundation models. It follows your permissions, sensitivity labels and retention policies. Web search queries go to Bing, stripped of identifiers, under separate terms. |
| Claude Team or Enterprise (Anthropic) | Reading long documents and careful drafting: a full set of financial statements, a trust deed, a long ruling. | Anthropic doesn't use commercial plan inputs or outputs to train its models by default. A chat a user rates with the thumbs up or down can be used for training, and plan owners can switch rating off. |
| JAX in Xero | Answers questions about the business's Xero data, and categorises and matches high-confidence bank transactions automatically. Xero says JAX chat is available to all subscribers at no extra charge for now. | Xero says JAX chat doesn't use your data to train the models it relies on, and that its model providers process the data but don't retain it or train on it. JAX follows each user's Xero permissions. |
| MYOB AI | AI BAS, in beta for MYOB Lite and Pro customers in Australia, prepares the BAS as you go, checks GST and flags anomalies. Smart Reconciliation and AI Business Insights are also in beta. | MYOB says the customer stays responsible for decisions made on AI suggestions, and recommends a trusted adviser reviews the output. It says it is ISO 27001 certified. |
| Karbon AI (practice management) | Summarises email threads, drafts and refines client emails, and builds a client brief from emails, notes, work and billing. | Karbon says firm data stays within Karbon and isn't used to train AI models. The features are on for every user by default, and an administrator can turn them off. |
Each entry is what the vendor's own page said on 30 September 2026. None of it is an endorsement. Features in beta change quickly, so check before you rely on one. See the sources.
Two things follow from the table. The data commitments belong to business plans: a staff member's personal account sits under consumer terms, and the firm can't see what goes into it. And platform AI can be switched on before anyone has decided to use it, so it belongs in the same rule as the assistants.
For how the three general assistants compare, see ChatGPT vs Copilot vs Claude. The usual answer is the one your firm already licenses. If you live in Microsoft 365, that is often Copilot.
Your professional obligations, and what they mean in practice
General information, not legal, tax or professional advice. Check how these rules apply to your firm with your professional body or your own adviser.
Neither the TPB nor the APESB has written new rules for AI. Their existing codes apply to it, and both have now said how.
Confidentiality
APES 110, the ethics code for members of CA ANZ, CPA Australia and the IPA, lets a member use or disclose confidential information only where there is a legal or professional duty or right to, or where the client has authorised it. The code, as revised for technology from 1 January 2025, says protecting that information covers its collection, use, transfer, storage, dissemination and destruction. A prompt touches most of those.
For registered tax and BAS agents, Code item 6 is blunter: "Unless you have a legal duty to do so, you must not disclose any information relating to a client's affairs to a third party without your client's permission." The TPB's guidance statement TPB(GS) 55/2026, issued on 22 July 2026, applies that to AI. You must get each client's permission before their information goes to a third party, and that "can include entering client information into AI models and tools, depending on how these tools are configured and used."
The permission can sit in a signed engagement letter, a signed consent or a general authority. The TPB recommends telling the client to whom and where the disclosure will be made, where the data will be stored and whether AI tools may be used. Tax file numbers bring in the Privacy (Tax File Number) Rule 2015 as well.
Competence and verification
The TPB says practitioners stay accountable for the accuracy of what they give clients. They should verify AI output at each step of the workflow and document those steps, which puts the checking into the records and quality management system the TPB's rules already require. Whether you took reasonable care depends partly on whether you reviewed the output before relying on it.
APES 110 asks every member to keep up with technology developments relevant to their work. The APESB's October 2025 technical alert says members "must have an inquiring mind to critically assess and verify any information used or produced for a professional activity, which includes AI-generated information."
Supervision
APES 110 requires a member to take reasonable steps to make sure the people working under them have appropriate training and supervision. The TPB's determination requires the same of anyone providing tax agent services on your behalf, inside a documented system of quality management. A junior drafting a client letter with AI is a supervision question, and the reviewer needs to know AI was used.
Disclosure
The APESB's technical alert says members "should disclose when AI tools are used and supervise and review their use." For valuation, forensic accounting and corporate finance reports, its service standards already require the methodology and sources to be disclosed, and the alert says that would generally include AI tools and AI-generated information. The APESB also began reviewing all its pronouncements for technology and AI in September 2025, so more guidance may follow.
Put that beside the TPB's permission rule and the practical answer is an engagement letter that says which AI tools the firm uses, for what, and where the data is held.
Privacy
Where the Privacy Act applies to your firm, putting a client's personal information into an AI tool is a use or disclosure that has to fit Australian Privacy Principle 6. The OAIC recommends, as best practice, that organisations don't enter personal information, and particularly sensitive information, into publicly available generative AI tools. It also expects due diligence on a product before it is adopted, and regular reviews after.
The risks, and the controls that deal with them
Most of what goes wrong is ordinary: a number nobody checked, or client data in a place the engagement letter never mentioned. The controls are ordinary too.
| Risk | How it shows up | The control |
|---|---|---|
| A wrong figure | A total, a rate or a date that reads as right and isn't. The tool states it with the same confidence as a correct one. | AI never originates a number that leaves the firm. The reviewer ties every figure back to the ledger, the return or the source document. |
| An invented reference | A ruling, a section or a case that doesn't exist, or doesn't say what the summary claims. | Cite the source, never the summary. Open the ruling or the legislation and read the passage before it goes into advice. |
| Client data leaving the firm | Someone pastes a client's details into a personal account, or a tool sends data somewhere the engagement letter doesn't cover. | Business plans only, with SSO and retention set before anyone starts. Feedback sharing off where the plan allows. AI named in the engagement letter. |
| Tax file numbers | A TFN inside a pasted return, an uploaded document or a prompt. | Keep TFNs out of general assistants. The TPB points to the Privacy (Tax File Number) Rule 2015 and says to take your own advice on how it applies. |
| AI already switched on | AI arrives inside software the firm already pays for, sometimes on by default. Karbon's is, for one. | List the platform AI in the same rule as the assistants, and set who can use it through roles and permissions. |
| A polished draft from a junior | An AI-assisted draft reaches the reviewer looking finished, and gets less scrutiny than a rough one would. | Note on the file when AI was used. The reviewer checks it the way they would check a first-year's work. |
A rollout that holds up
The risk to manage first is staff using personal accounts because the firm hasn't given them an approved tool. Banning AI doesn't fix that. Closing the gap does, in this order.
- Choose the approved tools and plans: one general assistant on a business plan, plus the AI already in your ledger and practice software. Set up SSO, retention and roles before anyone is invited, and turn off feedback sharing where the plan lets you.
- Update the engagement letter. Say which AI tools the firm uses, for what, and where client data is stored, and get the client's permission in the same document.
- Write the one-page rule: what may go into a prompt, what may not (tax file numbers, for a start), and who to ask when it is unclear. Our one-page AI policy template is a starting point.
- Train on the firm's own documents, such as last year's advice letters and a real reconciliation, with client details taken out. Invented examples teach people to use AI on invented work.
- Set the review rule and record it. Who checks AI-assisted work, what they check it against, and a line on the file saying AI was used.
Then start with one recurring task that takes real time, such as the covering letters at tax time or the monthly commentary for a management accounts client. Time it now, time it again once the team is using the tool, and let your own number decide what comes next.
How AIRE helps accounting firms
In an accounting firm the work product is a document: a return, a letter of advice, a set of management accounts, a file note. So our training runs on those documents. It also settles what client material may go into which tool, and who checks the output before it leaves the firm.
Our AI training for business teams is hands-on and live, on whichever assistant you already license: ChatGPT, Claude or Microsoft Copilot. Team workshops start from $5,500 + GST for groups of 5 or more. Across our training, 96% of participants said they were likely to apply what they learned (Post-training survey, n=220, October 2025 to September 2026).
For the rule itself, our AI strategy consulting and governance work sets which plan to run on each tool, with settings your IT provider will sign off, and writes the one-page data-handling rule. Advisory work starts with a 90-minute AI use case session at $2,900 + GST.
Working out which tools your firm should approve? The discovery call is free and takes 30 minutes.
Frequently asked questions
Can accountants use ChatGPT with client data?
Not on a personal account. The OAIC recommends against entering personal information into publicly available generative AI tools, and an article in CPA Australia's INPRACTICE quotes a member of its ethics and professional standards centre of excellence saying client data shouldn't go into a public-facing tool like ChatGPT. On ChatGPT Business, OpenAI says it doesn't train on your data by default. The Tax Practitioners Board still expects a client's permission before their information goes to a third party, and says entering it into an AI tool can count, depending on how the tool is configured and used. So put AI in the engagement letter and set a rule on what staff may paste in. This is general information, not professional advice.
Do we have to tell clients we use AI?
In practice, yes. The APESB's October 2025 technical alert says members should disclose when AI tools are used. For registered tax and BAS agents, the TPB says you need the client's permission before disclosing their information to a third party, that entering it into an AI tool can count, and recommends telling the client where the data will be stored and whether AI tools may be used. A signed engagement letter can carry that permission, so that is the natural place to say it. This is general information, not professional advice.
What does the Tax Practitioners Board say about AI?
Its guidance statement TPB(GS) 55/2026, issued on 22 July 2026, explains how the existing Code of Professional Conduct applies to AI. It adds no new obligations. Practitioners stay accountable for the accuracy of their advice, should verify AI output at each step of the workflow and document it, must get a client's permission before client information goes to a third party (which can include an AI tool), and should check that the AI tools they use keep information secure and meet the Privacy Act.
Can AI prepare a BAS or a tax return?
It can prepare parts of one. MYOB's AI BAS, in beta for its Lite and Pro customers, prepares the BAS as transactions come in, checks GST and flags anomalies, and MYOB recommends a trusted adviser reviews the output. Responsibility doesn't move. The TPB says practitioners remain responsible for the tax agent services they provide, and whether they took reasonable care depends partly on whether they checked the AI's output before relying on it.
Which AI tool should an accounting firm use?
Usually one general assistant on a business plan, plus the AI already inside your ledger and practice software. If the firm runs on Microsoft 365, Microsoft Copilot works inside Word, Excel and Outlook on files you already hold. ChatGPT Business and Claude Team are separate workspaces that people use for drafting and reading long documents. Whichever you choose, pick a plan whose terms say your data isn't used for training, and set it up before anyone uses it on client work.
Does Xero's JAX use client data to train AI?
Xero says it doesn't. Its JAX page says JAX chat doesn't use your data to train the large language models it relies on, that providers such as AWS, Microsoft Azure and GCP process the data but don't retain it or train on it, and that JAX follows each user's Xero permissions. Your own obligations still apply to what you ask it and to what you do with the answer.
Does AIRE train accounting firms?
Yes. We train accounting teams on ChatGPT, Claude, Microsoft Copilot and Gemini, on the firm's own documents, and settle what client material may go into which tool and who checks the output before it leaves the firm.
Sources
Every rule and product detail on this page comes from the publisher's own page, and we read each source on 30 September 2026. Guidance and features change, so check the current version before you rely on one. This guide is general information, not legal, tax or professional advice.
- Tax Practitioners Board, TPB(GS) 55/2026 The use of Artificial Intelligence and the Code of Professional Conduct (issued 22 July 2026). Read 30 September 2026.
- Accounting Professional and Ethical Standards Board, Technical Alert: The ethical use of artificial intelligence by professional accountants (31 October 2025). Read 30 September 2026.
- Accounting Professional and Ethical Standards Board, Technology-related revisions to APES 110 Code of Ethics for Professional Accountants (including Independence Standards) (issued June 2024, effective 1 January 2025). Read 30 September 2026.
- Accounting Professional and Ethical Standards Board, Technology: ethics and independence considerations. Read 30 September 2026.
- CPA Australia, How far can you go with AI before you hit an ethical dilemma? (INPRACTICE, 25 August 2025). Read 30 September 2026.
- Office of the Australian Information Commissioner, Guidance on privacy and the use of commercially available AI products (updated 17 January 2025). Read 30 September 2026.
- Xero, Just Ask Xero (JAX). Read 30 September 2026.
- MYOB, MYOB AI. Read 30 September 2026.
- Karbon, AI-powered practice management for accounting firms. Read 30 September 2026.
- OpenAI, Enterprise privacy at OpenAI (updated 8 January 2026). Read 30 September 2026.
- Microsoft Learn, Enterprise data protection in Microsoft Copilot and Microsoft Copilot Chat (last updated 18 August 2026). Read 30 September 2026.
- Anthropic Privacy Center, Is my data used for model training? (commercial products, dated 18 August 2026). Read 30 September 2026.
Related guides
- AI for law firms in Australia: the tools and the rules
Where AI helps an Australian law firm, which legal and general tools fit, and what the conduct rules and the courts' practice notes on AI require.
- A one-page AI policy template for Australian organisations
Copy a one-page AI policy for Australian organisations: approved tools, what data may go in, who checks output, who owns AI risk, and how to roll it out.
Related services
- AI training for business teams: hands-on sessions on the tools your team already licenses, run on your own documents.
- AI strategy consulting and governance: which plan to run on each tool, and a one-page rule on what may go into a prompt.