In a hypothetical practice, a tax senior pastes a client’s HMRC letter into a public chatbot, and a bookkeeper does the same with management accounts. Both take two minutes, which is the appeal.
On a public chatbot, the profession’s guidance points one way. On a paid tool it is less clear. This article argues that a useful first question is what the client would expect. This article is not legal advice.
ICAEW’s Code of Ethics bars disclosing confidential information (R114.2) without the client’s authority or a legal or professional right or duty (R114.3), and that covers staff (1.4A1).
Tax guidance under Professional Conduct in Relation to Taxation (PCRT) says at paragraph 4.2: “The input of client data into publicly available AI tools is likely to constitute a breach of client confidentiality, unless the client has consented to this.” ICAEW’s dos and don’ts reach the bookkeeper too: keep client data off public AI tools.
The PCRT guidance suggests tools “not made for public use” without saying whether a paid plan counts. An ICAEW article of 28 July 2026 wants “appropriate due diligence, contractual protections and internal approval” first. One ICAEW guide page, a contributor’s view, says organisations “should not put confidential information or personal data into a Generative AI tool”, then suggests a secure version.
Nothing read here says whether those steps are enough, or whether a contracted tool is a disclosure needing the client’s authority. ICAEW members can ask its confidential Ethics Advisory Service. Members of other bodies should read their own code, which this article has not.
The 28 July article suggests asking whether the client “would reasonably expect” the information to be used that way. It was offered for information moving inside a firm, so applying it to AI is this article’s argument, not ICAEW’s. A no suggests authority is missing. A yes does not supply it.
A contracted cloud tool may be able to meet those three steps. Once the question above is answered, and any authority is in place, it may be the practical route for many practices.
Your own premises is a choice, not a duty: no text read here requires it. It suits a firm that wants no outside supplier holding client data and can look after the machine. The client’s expectation applies there too. Local Assistant provides private, local AI for business, so weigh this paragraph accordingly.
Before the next paste
- Find out which AI tools staff already use.
- For each, check what the 28 July article says firms should understand: where data is stored, who can access it, whether it may train models, and how outputs are monitored.
- Check authority: PCRT guidance suggests reviewing engagement terms, without calling them consent.
- Personal data is a further question: see the ICO’s guidance on AI.
The client has read none of this and already has an expectation about where the file goes. What a practice controls is whether anyone there knows what it is.
Drafted with AI assistance and reviewed by Ilja Orlovs before publication.