AI legal advice tools risk waiving client privilege, courts warn

An omnipresent challenge for all solicitors, regardless of practice area, is being second-guessed by clients.

Family lawyers in particular will be familiar with the client who has a friend who has been through a divorce and is only too happy to pick apart every step the solicitor takes, often noting that the friend’s solicitor did things differently, and, inevitably, better.

The advent of the internet exacerbated this problem, giving clients ready access to articles, commentary and opinion from sources of varying levels of expertise, which they can use to challenge a practitioner’s advice and strategy.

That tendency has now been super-charged by the rise of tools based on large language models, which clients can interrogate for instant, and highly convincing second opinions. It is essential that practitioners actively discourage this practice, not only because of the well-documented inaccuracies of these tools, but also because such interactions are unlikely to be protected by professional privilege.

The risks were illustrated in the recent US case of United States v Heppner  (S.D.N.Y., 17 February 2026). Heppner was arrested on multiple charges linked to alleged misconduct as an executive of several corporate entities. A search warrant executed at his home uncovered documents that recording communications he had made with a generative AI platform.

Those communications occurred after Heppner became aware he was the target of an investigation. The documents set out defence strategies and potential arguments in response to charges he believed he might face. He prepared them without consulting his lawyers.

Heppner argued the documents were privileged, on the basis that:

• He had input information he learned from counsel into the platform;
• He had created the documents for the purpose of seeking advice from counsel; and
• He had subsequently shared the documents with counsel.

The court rejected that claim. It found:

• The documents were not communications between counsel and client – the AI platform was not a lawyer;
• The communications with the platform were not confidential, largely because the terms of use included permission for the platform’s owners to disclose the information to third parties (including the government);
• Heppner was not seeking legal advice from the platform, despite the fact that his purpose in communicating with the platform was talking to counsel, because counsel had not directed him to use the platform. Sharing the communications with counsel at a later date did not imbue them with privilege.

While this is a US decision, the same outcome would likely follow under Australian law. It is well established that privilege can be lost if a client behaves inconsistently with its maintenance.¹

In addition, AI platforms of the kind used in Heppner almost invariably operate under terms incompatible with confidentiality, and later involvement of a lawyer cannot convert unprivileged material privileged communications.

In view of this, client onboarding, particularly for matters that may lead to litigation, should include a clear warning against the use of any AI-based tools or websites to discuss the substance of their case. Clients need to understand the real risks of waiving privilege in doing so.

Practitioners should also ask clients whether they have already used such tools, to avoid unpleasant surprises later.

Finally, even with appropriate warnings, clients may still be tempted to seek reassurance from AI tools if they feel under‑informed about their matter. As so often, clear, regular and effective communication remains the best safeguard.

Footnotes
¹ Mastercard Asia/Pacific (Australia) Pty Ltd v Australian Competition and Consumer Commission [2026] FCAFC 37.

The Queensland Law Society has drafted a warning for clients template.