AI agents for accounting practices: what the TPB's 2026 AI guidance allows and where automation pays

Key takeaways

• The TPB’s AI guidance, TPB(GS) 55/2026, is now final. Practitioners remain responsible for anything AI produces.
• Client permission comes before client information goes into an AI tool that discloses it to a third party.
• AI agents for accounting practices fit the front office: enquiries, bookings, chasers and reminders.
• Advice, judgement and lodgment decisions stay with registered practitioners.

What does the TPB AI guidance actually say?

The Tax Practitioners Board released TPB(GS) 55/2026 in July 2026, after an exposure draft in March (TPB announcement). Three points matter most for a small practice:

1. Responsibility does not move. Practitioners “are still ultimately responsible” for the services they provide when they use AI (paragraph 8).
2. Check the output. Practitioners must verify and review AI-generated content (paragraph 16).
3. Ask before sharing. Practitioners must obtain permission from each client before divulging client information to a third party, which can include entering it into AI models and tools (paragraph 23).

The TPB also publishes a one-page factsheet that is worth sharing with your team.

How do AI and client confidentiality fit together?

The Code’s confidentiality obligation already covers third parties such as cloud storage, and the TPB suggests telling clients about those arrangements (TPB). AI and client confidentiality work the same way: know where client information goes, have permission where it is disclosed, and tell clients in your engagement letter.

CPA Australia put it plainly in August 2025: uploading client data into public AI tools such as ChatGPT is not OK (CPA Australia).

A practical line for a small practice:

• Fine without client data: drafting a generic newsletter, summarising a public ATO update, writing a reminder template.
• Needs permission and a vetted tool: anything that processes a named client’s financial or tax information.
• Not for AI: final advice, positions taken in a return, anything you would sign.

Where does accounting practice automation pay off first?

The front office, where the work is repetitive and the rules are clear. Accounting practice automation that is low risk and high value:

Job	What the agent does	Client data involved
New enquiries	Answers service and fee questions, books a first meeting	Name and contact only
Onboarding	Sends the checklist, chases missing items, records the ID check	Held in your own systems
Keeping clients informed	Sends the section 45 disclosures with every engagement	None beyond the contact
Lodgment reminders	Sends reminders before each program date with the document list	Name and due date
Payment	Sends proposals, invoices and payment links	Billing details in the CRM

The TPB’s keeping clients informed obligation, for example, requires five items in writing to every current and prospective client, which is exactly the kind of task automation does reliably. Our guide on TPB keeping clients informed lists them.

For practices providing designated services under AML/CTF tranche 2, obligations started on 1 July 2026 and enrolment with AUSTRAC was due by 29 July 2026, or within 28 days of first providing a designated service (AUSTRAC). Onboarding agents can collect the documents; the due diligence decision stays with a person. See tranche 2 client onboarding.

What should an AI receptionist for accountants do?

An AI receptionist for accountants, by voice or chat, should:

1. Introduce itself as the practice’s AI assistant.
2. Answer general questions about services, location and the fees you are happy to publish.
3. Capture the enquiry type, such as individual return, business, SMSF or bookkeeping.
4. Book a first meeting with the right person.
5. Send the document checklist for that meeting.
6. Never give tax advice or comment on someone’s situation.

In peak season this is the difference between a full voicemail box and a calendar of booked first meetings.

What should accounting practice software with AI agents include?

Accounting practice software for the front office should connect to the tools you already run, rather than replace them. LYD Accrual OS publishes integrations with Xero, MYOB, QuickBooks, Karbon, Ignition, Microsoft 365 and Google Workspace, and covers enquiry capture, booking, onboarding, proposals, invoices, payment links and appointment reminders.

In October 2026 its published packages started from A$295 a month plus GST, with five users included. Check the current packages before you decide, and confirm in writing where client data is stored and who can access it.

Who builds this for accounting practices?

Launch Your Dream, by Sharktech Global, builds LYD Accrual OS for Australian accounting and bookkeeping practices, up and running within 7 working days of receiving your onboarding details. Read about agentic AI for small business and managed AI services, open the live demo or book a free 30 minute discovery call.

General information only, not legal or professional advice. Read TPB(GS) 55/2026 and check the TPB and AUSTRAC for current requirements.