← Resources

TPB's draft AI guidance (D62/2026): what it actually asks of your practice

19 July 2026AccuMate

In March 2026 the Tax Practitioners Board released Exposure Draft TPB(I) D62/2026: The use of artificial intelligence and the Code of Professional Conduct. It is the TPB’s most direct statement yet on what the Code expects when a registered tax or BAS agent brings AI into their workflow. Public consultation closed on 21 April 2026, and the guidance remains in draft while the TPB works through submissions.

If you run an Australian practice, here is the short version.

The core principle: AI never carries your obligations

The draft’s central message is simple: using AI does not reduce or transfer a practitioner’s professional responsibilities. You remain fully accountable for the accuracy and quality of every service you provide, whether a person drafted it, software drafted it, or an AI model drafted it.

That means AI output is a draft input to your professional judgement, not a substitute for it. If an AI tool prepares a reconciliation, a BAS figure, or client advice, the Code expects the same review and professional care you would apply to work from a junior team member.

The Code obligations the draft highlights

The draft walks through how existing Code obligations apply to AI use, with particular attention to:

  • Confidentiality of client information. Client data doesn’t stop being confidential because it enters an AI tool. Practitioners need to understand where client information goes, who can access it, and whether the tool uses it for anything beyond serving the client.
  • Independence. Relying on a tool doesn’t outsource your independent judgement.
  • Honesty and integrity. The same standards apply to AI-assisted work as to any other work product.

What to do now, while it’s still a draft

The guidance is not yet final, but its direction is clear and consistent with how the TPB already frames technology use. Sensible steps today:

  1. Keep a human review step on every AI-assisted output. If your workflow has a path where AI-generated work reaches a client or the ATO without practitioner review, close it.
  2. Know your tools’ data handling. Where is client data processed and stored? Is it used to train models that serve other customers? If a vendor can’t answer plainly, that’s your answer.
  3. Be transparent with clients about how their information is handled when AI tools are involved. Transparency is the safest reading of the confidentiality obligation, whatever final form the guidance takes.
  4. Document your review. The practitioner sign-off that the Code expects is easier to demonstrate when it’s recorded, not implicit.

Where AccuMate stands

AccuMate is built to this standard rather than retrofitted to it: every lodgement and client output passes a named practitioner sign-off gate, client consent is captured at onboarding, actions are audit-trailed, and client data is never used to train general models. The draft guidance describes the architecture we started with. See Security & Trust for the full regulation-to-feature map.

We’ll update this explainer when the TPB finalises the guidance.

Sources

This explainer is general information, not advice. Check the TPB’s current guidance for your obligations.