Does Putting Client Data Into AI Tools Violate Professional Confidentiality or Privilege Obligations?

Detailed Explanation

General data-safety guidance — classify the data, check the AI vendor's plan and terms, confirm who's the data controller — covers most business use cases well. Regulated professions carry an additional layer on top of that: a duty owed specifically to the client, patient, or beneficiary, which can be stricter than general data-protection law and doesn't disappear just because a vendor's terms look otherwise acceptable.

Legal. Legal professional privilege protects certain communications from disclosure, and a solicitor's separate duty of confidentiality — set out in the Legal Profession Uniform Law's Australian Solicitors' Conduct Rules (rule 9.1) — covers a broader set of client information regardless of privilege status. The Law Society of NSW, the Legal Practice Board of WA, the Victorian Legal Services Board and Commissioner, and equivalent bodies in other states have issued guidance on generative AI use — the emerging consensus is that AI tools can be used with client information, but generally require informed client consent for anything sensitive, a vendor whose terms don't permit training on submitted data, and continued independent professional judgment rather than treating AI output as a substitute for legal analysis. Courts, including several Australian courts, have also begun issuing practice notes requiring disclosure of AI use in prepared materials.

Healthcare. Health information carries heightened protection under the Privacy Act 1988 as a form of sensitive information, with extra obligations layered on top for organisations connected to the My Health Records Act 2012 system, and professional obligations from AHPRA-registered practitioners' codes of conduct. Using a general-purpose AI tool with patient information typically requires a vendor agreement with safeguards that meet this higher bar — not just a standard commercial data-handling agreement — and many general-purpose AI products aren't configured or contracted for this by default.

Financial services. Financial advisors, accountants, and similar roles carry fiduciary and confidentiality duties to clients, often layered with ASIC-regulated obligations (for financial advisors) or professional-body standards (for accountants) around client financial data. As with law and healthcare, the question isn't only "is this AI vendor generally safe with company data" — it's whether the specific professional and regulatory duty owed to this particular client has been satisfied.

The through-line across all three: general data-safety practice (see is it safe to put company data into AI tools) is necessary but not sufficient. A profession-specific duty can require more — informed consent, a specific type of vendor agreement, or in some cases, treating certain material as unsuitable for any general-purpose AI tool at all.

What to Actually Do

1. Identify which professional duty applies before assessing any specific AI use. A law firm, a medical practice, and a financial advisory business each answer to different rules — check your specific professional body's current guidance rather than assuming general data-protection compliance covers you.

2. Treat informed client consent as the default requirement, not an edge case. Where guidance calls for it, get consent before routine use, not only when a client happens to ask — building this into standard engagement or intake paperwork is more reliable than relying on staff to remember case by case.

3. Vet the AI vendor specifically for the professional standard, not just general business terms. A standard data processing agreement may not satisfy a healthcare-specific business-associate requirement, for example — confirm the vendor offers the specific safeguards your profession's guidance calls for, not just a generic enterprise privacy policy.

4. Decide, in writing, what categories of material never go into a general-purpose AI tool. Some information — highly sensitive case strategy, certain categories of health data, specific client financial details — may warrant a blanket rule rather than a case-by-case judgment call, especially for a business without in-house counsel to make that call consistently.

5. Build this into the same usage policy that covers general AI use. See what should an employee AI usage policy include — a regulated-profession business should add explicit, profession-specific rules on top of the general policy, not assume the general version implicitly covers this.

Things to Consider

• This compounds with, rather than replaces, general data-safety practice. Both layers apply — see is it safe to put company data into AI tools for the foundational classification and vendor-vetting framework this builds on top of.
• Contract and document review is one specific case of this broader question. See how do you use AI assistants to review contracts and legal documents for the practical workflow considerations, and treat confidentiality-clause and privilege questions as the compliance layer underneath that workflow.
• Vendor due diligence needs the same rigor as any other tool decision, arguably more. See how do you evaluate an AI vendor's data processing agreement before adopting a tool — a regulated professional business should treat this evaluation as a prerequisite, not an afterthought, before any client information reaches the tool.
• Guidance in this area is genuinely still evolving. Professional bodies are actively updating their positions as generative AI use becomes more common — treat any specific rule cited here as needing verification against your professional body's current guidance, not a fixed, permanent answer.

Common Mistakes

• Assuming a general data-safety check ("we're on a business plan with a DPA") automatically satisfies a professional confidentiality or privilege duty. The two are related but distinct — a vendor being generally trustworthy with business data doesn't establish that a specific professional-duty bar has been cleared.
• Skipping informed client consent because the AI use feels routine. Guidance in most professions treats consent as something to obtain proactively, not something that only matters for unusual or high-stakes uses.
• Treating anonymization as a complete solution without checking whether it actually anonymizes. A redacted document with enough remaining case-specific detail, or a small client base where indirect identifiers still point to one person, can remain effectively identifiable even with names removed.
• Leaving this out of the AI usage policy because it feels like "the lawyer's problem" or "the doctor's problem" rather than a business-wide policy issue. Every staff member handling client-facing work, not just the licensed professional, needs to understand and follow the same rule.