Do You Need a Client's Permission Before Putting Their Tax File Into AI?

Detailed Explanation

Tax and BAS agents have started using AI tools to help extract data from receipts and statements, draft client correspondence, summarise financial records, and prepare working papers — often without a clear answer to a basic question: does putting a client's tax file into an AI tool require telling the client first?

The Tax Practitioners Board answered this directly in TPB(GS) 55/2026, guidance interpreting how the existing Code of Professional Conduct applies to AI use. The guidance doesn't create a brand-new rule out of nothing — it applies confidentiality obligations that already exist under the Code to a genuinely new kind of third party: an AI vendor processing the client's information.

Why Putting Data Into AI Counts as a Disclosure

Code item 6 is in section 30-10 of the Tax Agent Services Act 2009. It prohibits disclosure of information about a client's affairs to a third party without the client's permission, unless there is a legal duty to disclose it. The TPB says entering information into an AI tool can be a third-party disclosure, depending on how that tool is configured and used.

Where the proposed AI use involves such a disclosure, obtain the client's permission first, through an engagement letter, signed consent or another clear communication. The TPB recommends explaining to whom and where information will be disclosed, where it will be stored and whether AI may be used. That explanation is a recommendation in the guidance; the underlying permission requirement is the legal obligation.

The Documented-Review Requirement

Consent isn't the only obligation the guidance sets out. It also expects a practitioner to verify and review AI output at each step of the workflow, with each of those steps documented. This connects the AI-specific guidance back to two existing obligations under the 2024 Code Determination: section 30 (keeping proper client records) and section 40 (maintaining quality management systems) — both of which the TPB frames AI-assisted work as needing to satisfy in the same way as any other client work.

In practice, this means a practitioner using AI at multiple points in preparing a return — extracting figures, drafting sections, checking calculations — should be able to show that each of those points was reviewed by a person, not just that the finished return looked right at the end. A single sign-off on the final output is a weaker position than a documented check at each stage where AI touched the work.

The guidance is also explicit that AI "cannot be relied on as a replacement for tax knowledge, experience or expertise" — a direct statement that AI assistance doesn't lower the professional judgment bar a registered agent is expected to meet.

Tax File Numbers Specifically

Where a client's tax file number itself is part of what's being processed, an extra layer applies. The Privacy (Tax File Number) Rule 2015 sets specific handling obligations for TFN information, on top of the general Privacy Act obligations that apply to personal information more broadly. A practitioner putting TFN-containing documents into an AI tool should treat that as carrying the heightened handling standard the TFN Rule sets, not just the general confidentiality duty under the Code.

Where This Sits Relative to the Code's Staged Commencement

The Code Determination's staged commencement is now complete for every practitioner: firms with more than 100 employees were bound from 1 January 2025, and firms with 100 or fewer employees from 1 July 2025. This means every registered tax and BAS agent, regardless of firm size, is now subject to the full set of obligations TPB(GS) 55/2026 interprets — this isn't a future requirement building up to a deadline, it's a current one.

What a Practical Response Looks Like

For a small or mid-sized practice, meeting this obligation doesn't need to be elaborate:

1. Add AI use to the engagement letter. A clause explaining that the practice may use AI tools to assist with parts of the client's work, naming (in general terms) what kind of assistance that involves and where data is processed, covers the consent requirement for future engagements without needing a separate conversation for every task.
2. Get explicit consent for existing clients, not just new ones. A short, clear communication to existing clients — an email or a note at next contact — brings existing engagements into the same position as new ones, rather than leaving a gap for clients who signed on before AI use began.
3. Build a habit of recording each review step, not just the final check. This doesn't need special software — a short note against each stage of a workflow (who reviewed the AI-extracted figures, who checked the draft return) is enough, provided it happens consistently. See what should you log for every AI-assisted task so you can explain it later for the general version of this discipline.
4. Explain the proposed disclosure clearly. The TPB recommends telling clients who will receive the information, where it will be disclosed and stored, and whether AI may be used. Check the actual vendor arrangement before describing it to clients.

Things to Consider

• This sits on top of existing document-handling practices, not instead of them. A practice already collecting client documents through a secure portal (see how do accounting and bookkeeping firms automate client document collection) should extend that same discipline to cover AI-assisted processing of what's collected, rather than treating AI use as a separate, unaddressed step.
• A general AI usage policy doesn't automatically cover client consent. An internal policy governing how staff use AI tools is a different document from client-facing consent — a practice needs both, and one doesn't substitute for the other.
• This is a live, current obligation, not a future one. With the Code's staged commencement now complete for all practice sizes, there's no grace period remaining to point to — the obligation applies now.
• Revisit consent language as AI use in the practice expands. A consent clause written when AI was used only for document summarisation may not adequately describe a practice that has since started using AI to draft return sections — keep the client-facing description current with actual use.

Common Mistakes

• Assuming an existing general privacy policy already covers this. A standard privacy policy addressing data collection broadly often doesn't specifically address AI tool use as a third-party disclosure requiring its own consent — check the specific wording rather than assuming coverage.
• Getting consent once and never revisiting it as AI use grows. A practice that expands what it uses AI for over time should revisit whether its original consent language still accurately describes current use.
• Treating a single end-of-process review as sufficient documentation. The guidance's framing is step-by-step verification, not one review at the end — a single sign-off doesn't demonstrate the same level of oversight as a documented check at each stage.
• Overlooking the TFN Rule's extra handling requirement when tax file numbers are specifically involved. General confidentiality practice under the Code doesn't automatically satisfy the Privacy (Tax File Number) Rule 2015's more specific requirements — treat TFN-containing data with the heightened standard it requires.