How to safely connect ATO data to the AI tools your firm already uses

The privacy regulator has drawn a line: do not enter personal information into publicly available AI tools.

For firms pasting client ATO data into Claude, ChatGPT or Copilot to speed up research, cross-client analysis or client meeting prep, that is the position to work from. The OAIC is not restricting the use of AI with client data. It is restricting where the data goes and how.

The TPB has since confirmed that the same logic applies in tax practice. Together, the two documents set a standard most firms have not yet met.

The OAIC published its Guidance on privacy and the use of commercially available AI products in October 2024 and updated it in January 2025. The guidance explains how the Australian Privacy Principles apply when a business runs client information through a commercial AI product.

The full OAIC position reads: “As a matter of best practice, the OAIC recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools, due to the significant and complex privacy risks involved.”

Four of the Australian Privacy Principles are the ones that bite hardest here.

APP 6 limits what personal information can be used for. The OAIC’s position is that “in cases of ambiguity, the primary purpose for collection, use or disclosure should be construed narrowly rather than expansively.” A client who handed over data to prepare a BAS did not necessarily hand it over to be processed through a US-hosted chatbot.

APP 11 requires reasonable security. The OAIC notes that “deploying a system locally or on premises is likely to be more privacy-preserving as it limits the risks of third party access to the data.”

APP 10 requires reasonable steps on accuracy. That is a harder standard to meet when the tool is probabilistic by design.

APP 5 requires that privacy policies and collection notices identify any AI-related purposes. Most firm privacy policies drafted before 2024 do not.

On 22 July 2026, the Tax Practitioners Board published TPB(GS) 55/2026: The use of Artificial Intelligence and the Code of Professional Conduct. It explains how the existing Code applies when AI is used in tax agent services.

Code item 6 of the Code prohibits a tax practitioner from disclosing information about a client’s affairs to a third party without the client’s permission. For years, firms have treated it as a rule about external contractors and offshore service providers. In July the TPB said it also applies to AI tools.

The exact wording matters: “In the context of AI in the provision of tax agent services, tax practitioners must obtain permission from each client prior to divulging client information to a third party (which can include entering client information into AI models and tools, depending on how these tools are configured and used).”

For firms whose engagement letters do not reference AI, the TPB has set an obligation. Permission needs to exist before the data goes in, not after.

Code item 10 then applies to the output. Reasonable care requires that AI output be verified before it is relied on. Section 40 of the Determination applies on top: every firm needs a documented system of quality management covering what is permitted, who is trained on it and how outputs are reviewed. The TPB also flags that where client information involves tax file numbers, additional obligations under the Privacy (Tax File Number) Rule 2015 apply. ATO data almost always includes TFNs.

Neither document says firms should stop. Both say there is a right way and a wrong way, and the regulators have landed in roughly the same place.

The firms moving fastest have treated the regulatory position as the starting point. Engagement letters cover AI use, so client permission exists before the first prompt. The tool has been reviewed before anyone uses it: where it is hosted, who can access inputs, whether the vendor’s terms permit the use the firm intends. What is permitted and what is not is written down, so the team is not making the call case by case. TFN data has been thought through specifically. Configurations keep client data within the firm’s control.

The time savings on the other side of that work are why firms bother. Research that used to take an afternoon runs in a conversation. Cross-client views that used to need a spreadsheet show up in a single prompt. Client meeting prep that used to take an hour is done in ten minutes.

The gap between the firms moving on this and the firms waiting is mostly about seeing what a working setup looks like. Stan Corner, Founder and CEO of Business Automation Works, is covering exactly that at The Firm’s AI in Practice Summit on 11 November. His session walks through how practices can securely connect client ATO data with ChatGPT, Microsoft Copilot and Claude. He covers the use cases that make the work worth it: surfacing overdue lodgments, debts and obligations, preparing client meetings with aggregated ATO information, and turning that data into clear client summaries and communications.

The Future of ATO Data + AI: Securely Connecting with ChatGPT, Copilot and Claude

Wednesday 11 November 2026, 11:00am AEDT

Register for the session

Part of The AI in Practice Summit, 11 and 12 November 2026. Free, online, every session recorded.