APES 110 confidentiality and AI: what Subsection 114 requires.

Of the five fundamental principles in APES 110, confidentiality is the one most AI decisions turn on. Every AI tool does something with information: reads it, sends it somewhere, keeps what it was asked. Subsection 114 governs all of that for client information, and its application material now describes confidentiality in terms that map closely onto how AI services handle data.

What Subsection 114 requires

R114.1 requires members to respect the confidentiality of information acquired in the course of professional and business relationships. It then lists specific duties, two of which matter most here: be alert to the possibility of inadvertent disclosure, and take reasonable steps to ensure that personnel under the member’s control, and individuals from whom advice and assistance are obtained, comply with the member’s duty of confidentiality.

R114.2 prohibits disclosing confidential information acquired in professional and business relationships, using it for the advantage of the member, the firm or a third party, and using or disclosing it after the relationship has ended. AUST R114.3, an Australian addition to the international code, sets out the exceptions: disclosure or use is permitted where there is a legal or professional duty or right to do so, or where the client has authorised it and it is not prohibited by law or regulation.

Why it reaches AI tools

Application material at 114.1 A1 says maintaining confidentiality involves taking appropriate action to protect information in the course of its collection, use, transfer, storage or retention, dissemination and lawful destruction. Hold a cloud AI tool against that list and each stage raises a question:

The authorisation route

If the firm wants to use a tool that sends client information to a third party, client authorisation under AUST R114.3 is the most defensible footing. The Code’s guidance on seeking authorisation for technology uses (114.3 A3) suggests communicating, preferably in writing, the nature of the information, the purpose, who will carry out the activity, and whether the client will be identifiable from the output. Engagement terms are the natural place for this; our page on engagement letters shows how firms draft them from their own templates.

The factors at 114.3 A2 are also worth reading against any AI vendor, in particular whether the recipients are appropriate, and which privacy laws apply in the jurisdiction where disclosure might take place. An offshore AI provider brings a second jurisdiction’s law into the picture.

Inadvertent disclosure is the everyday risk

Most confidentiality breaches involving AI will not be decisions. They will be a staff member at deadline pasting a client’s financials into a free chatbot because it is quick. R114.1 asks members to be alert to inadvertent disclosure and to take reasonable steps with the people under their control, which is where a written firm AI policy and training come in. The most reliable step is giving staff an approved tool that is quicker than the public one.

Where deployment posture fits

When AI runs on hardware inside the firm, the transfer, retention and dissemination questions above do not arise, because no client information goes to an AI provider. The remaining confidentiality work is what the firm already does for its file server: physical security, network security and access control. That is the case for an on-premise unit for accounting practices. The behavioural duties remain either way, including staff discipline about public tools.

Common questions