Is Pasting Client Data Into ChatGPT a Breach of Code Item 6?
Published 12 August 2026
Short answer: It depends on what you pasted, whether the client gave specific permission, and what that AI tool does with the input. But for most firms, most of the time, the honest answer right now is "probably, and we haven't checked." Under TPB(GS) 55/2026, an AI tool can be a third party for the purposes of Code Item 6.
This question has come up constantly since TPB(GS) 55/2026 was finalised in July 2026, and it deserves a straight answer rather than a vague "it's complicated."
The actual test
Code Item 6 bars disclosing a client's affairs to a third party without permission, unless there's a legal duty to disclose. The TPB has now confirmed that an AI tool can be a third party for this purpose. So the test is really two questions:
- Did the client give specific, documented permission for their information to be processed by an AI tool, knowing what tool, what data, and where it goes?
- If not, was there a legal duty to disclose it anyway? For most day-to-day AI use, no, there wasn't.
If the answer to both is no, pasting a client's bank statement or trust structure into a public AI chatbot is treated the same way unauthorised outsourcing would be. Not because the AI is malicious, but because nobody agreed to it going there.
Why "we didn't mean anything by it" doesn't help
A lot of AI use in accounting practices isn't a deliberate decision. It's a staff member trying to save time on a Friday afternoon, summarising a file or drafting a client email. The TPB's guidance doesn't carve out an exception for good intentions. The disclosure happens the moment the data leaves your systems, regardless of why.
This also catches AI features that are switched on by default inside tools you already use: practice management software, email clients, document platforms. If you haven't specifically checked whether those features send data externally, you may be disclosing client information you didn't consciously choose to disclose at all.
What actually fixes this
There are really only two categories of fix, and the strongest approach uses both:
- Get the paperwork right. Update engagement letters to capture specific, informed, documented consent, naming the categories of AI tool in use, what's shared, and where it goes.
- Change where the data can go in the first place. If your AI tools run entirely on hardware inside your own office, there's no third party for the data to be disclosed to. You still need the consent conversation for good practice, but the risk profile changes completely: no external vendor, no offshore server, no sub-processor chain to explain.
The second option doesn't remove your obligations under the Code. It removes the hardest, riskiest part of meeting them.
Not sure if your current AI use is covered?
Book a 20-minute AI and confidentiality audit, no obligation.
Book a 20-Minute AuditCommon questions
Is using ChatGPT with client data a breach of Code Item 6?
It can be. If a client hasn't given specific, documented permission for their information to go to that AI tool, and there's no legal duty to disclose it, entering it into a third-party AI tool can be an unauthorised disclosure under Code Item 6.
Does the TPB accept "we didn't mean to" as a defence?
No. TPB(GS) 55/2026 doesn't carve out an exception for good intentions. The disclosure happens when the data leaves your systems.
What's the fix?
Update engagement letters for specific, documented consent, and reduce where client data can go. AI that runs on hardware inside your office has no third party in the data path.