AI Compliance for Accounting Firms: What to Tell Clients

Here's the line most Australian accounting firms still haven't drawn. The moment a staff member pastes a client's financials into the free version of ChatGPT, the firm may have breached its confidentiality obligations under APES 110 — not because the output was wrong, but because the data left the building. And as of 22 July, the Tax Practitioners Board has closed the fallback: "the AI did it" is not a defence.

On 22 July 2026 the TPB published its finalised guidance, TPB(GS) 55/2026, on how the Code of Professional Conduct applies when tax practitioners use artificial intelligence. The core principle is short and non-negotiable: AI "does not replace professional judgement or transfer responsibility." You own every output — whether a person or a model produced it.

The guidance runs existing Code obligations straight through AI use: competence, reasonable care, confidentiality, record-keeping, professional judgement, and supervision. In practice that means you must vet an AI application for data security and Privacy Act 1988 compliance before it touches client data, and you must "obtain permission from each client prior to disclosing client information to a third party." A public AI tool is a third party. TPB Chair Peter de Cure AM framed the upside plainly — AI "has the potential to increase productivity, drive efficiencies and enhance client service" — but only "when used appropriately."

CPA Australia has been blunter. Brendan O'Connell, of its Centre of Excellence for Ethics and Professional Standards, puts it in one line: uploading "client data into public AI tools such as ChatGPT is not OK, as confidentiality is not guaranteed." In-house tools with proper privacy and security controls are fine. The consumer versions of ChatGPT, Claude and Gemini are not. This isn't a new rule invented for AI — APES 110 was amended from 1 January 2025 to spell out exactly these technology-and-confidentiality considerations. The guidance just applies it.

Want to know what this means for your business? We'll map it in one call.

Three obligations, three exposures. First, confidentiality. The biggest breach risk in most firms isn't a rogue decision to fire up ChatGPT — it's AI features already switched on inside Xero, MYOB and Microsoft 365, processing client data through models nobody consciously opted into. We've covered the AI already embedded in your accounting software and what the Code demands of it — that's where most firms are quietly exposed.

Second, accountability. As O'Connell puts it, you "can't just say to your client, I'm sorry, I got that from ChatGPT." Every AI-generated figure, letter or reconciliation has to be verified before it leaves the firm — the review step is a Code obligation, not a nice-to-have. This is the admin leverage of AI with a guardrail: the time it saves on drafting is real, but only if the verification discipline is real too.

If you're not certain where AI already touches client data across your practice — the embedded features and the free tools staff signed up for — map your firm's AI exposure before an auditor does. You can't write a policy for a surface you haven't measured.

Third, disclosure — the one most firms miss. CPA Australia recommends spelling out in your engagement letters "how you are using AI and how you're not using it." And from 10 December 2026, the Privacy Act adds a legal layer: regulated businesses must disclose in their privacy policy how personal information feeds substantially automated decisions — the automated-decision disclosure rule we've written about.

You don't need a governance framework to start. You need three things. Write a one-line firm policy: no client data in public AI tools, full stop, and name the sanctioned in-house or enterprise tools people may use instead. Add an AI clause to your engagement letters stating how you use AI and how you verify it. And run one audit of what's already switched on — the AI toggles in your practice software, the browser extensions on staff machines, the free tool someone signed up for last month. That last one is where the surprises live. For the full regulatory picture, our guide to Australia's AI compliance rules maps how the Code, the Privacy Act and Consumer Law fit together.

Key takeaways

Common questions

Related notes

Not sure where AI already touches client data in your firm?

A short conversation can map where AI already runs across your practice — the embedded features and the free tools — and turn it into a one-page compliance plan. Book a call.