Is the AI Already Inside Xero, MYOB or Microsoft 365 a Disclosure Risk?
Published 13 August 2026
Short answer: Possibly, yes, and it's the part of TPB(GS) 55/2026 most practices haven't checked. If a feature inside software you already use sends client data to an AI model to generate a summary, a draft, or a suggestion, that can be a disclosure to a third party under Code Item 6, whether or not anyone in the firm consciously decided to "use AI" that day.
Most conversations about AI and client data assume the risk is a staff member deliberately opening ChatGPT or Copilot and pasting something in. That's a real risk, and it's the easy one to spot. The harder one is already running quietly inside tools your firm uses every day.
Why this category of risk is so easy to miss
Nobody signs off on "using AI" when a feature is bundled into an existing subscription and switched on by default. Xero, MYOB, and Microsoft 365 have all been adding generative AI features (drafting assistance, smart summaries, predictive categorisation, meeting transcription) into products firms were already paying for. In most cases, these features arrived through a routine software update, not a decision anyone at the firm made.
That matters because TPB(GS) 55/2026 doesn't distinguish between "we decided to use an AI tool" and "an AI feature we didn't know about processed client data." The disclosure happens when client information leaves your systems and reaches a third-party model. The intent behind it is beside the point.
What to actually check
A basic audit doesn't need to be technical to be useful. Worth walking through with your practice manager or IT contact:
- Practice management software: does it offer AI-assisted drafting, summarisation, or search, and is it on by default?
- Email and document platforms: Microsoft 365 and Google Workspace both now ship generative AI features that can read across a mailbox or shared drive unless specifically disabled.
- Accounting platforms: Xero and MYOB's AI-assisted categorisation and reporting tools process transaction-level client data. Worth knowing exactly what that processing involves and where it happens.
- Anything labelled "smart," "assistant," "copilot," or "AI-powered" in a settings menu you haven't opened recently.
Why this is worse than the deliberate-use case, not better
It's tempting to assume embedded AI features are lower-risk because a vendor built them and presumably thought about compliance. That's not necessarily true, and even where it is, it doesn't solve your problem. TPB(GS) 55/2026 puts the disclosure obligation on the practice, not the software vendor. A well-designed AI feature from a reputable vendor is still a third party receiving client information if you don't have documented client permission for it.
There's also a practical problem. A deliberate ChatGPT session is something a principal can ask staff to stop doing. An AI feature quietly running inside core practice software has to be found before it can be addressed, and most firms haven't gone looking.
What to actually do about it
- Run the audit above: a one-off pass through your core software's settings, looking specifically for AI features and whether they're enabled.
- Decide, feature by feature, whether the processing needs specific client permission, and whether your current engagement letter actually covers it.
- Where a feature can be disabled and the practice doesn't need it, turn it off rather than leaving it running by default while you sort out the consent question.
None of this requires abandoning the software. It requires knowing what it's already doing. And where the answer to "where does this data go" needs to become simpler across the board, running your team's day-to-day AI assistance on hardware inside your own office removes this category of risk: there's no embedded-feature question to audit if the AI your team uses doesn't send data anywhere to begin with.
Not sure what's already switched on in your stack?
Book a 20-minute AI and confidentiality audit, no obligation.
Book a 20-Minute Audit