In an era where artificial intelligence tools are as accessible as a web browser, clients increasingly turn to AI for preliminary legal guidance before consulting their lawyers or to interpret legal advice they receive from lawyers. Whilst this might seem efficient and cost-effective, in a 2024 article we speculated use of these AI systems may have ramifications for legal professional privilege.

February 2026 US court decision serves as a stark illustration of these concerns, and a warning that use of AI could inadvertently waive legal professional privilege or generate materials which do not attract privilege protections in the first place.

What is privilege?

Legal professional privilege is one of the most fundamental protections in our legal system. It ensures that confidential communications between lawyers and clients remain protected from disclosure, enabling clients to speak freely and receive candid advice. In Australia, legal privilege is subdivided into advice privilege, which protects confidential communications between lawyer and client made for the purpose of seeking or giving legal advice, and litigation privilege, which protects documents created for the dominant purpose of use in existing or reasonably anticipated litigation.

A client is generally not required to disclose privileged communications in court proceedings, or in response to a regulatory notice. However, this privilege can be waived by acting in a way that is inconsistent with maintaining confidentiality over the privileged communication. For example, this can occur where a client either intentionally (or inadvertently) discloses or publishes the gist, substance or conclusion of a privileged communication to a third party.

Warning from US

A recent US court decision has rejected privilege claims over AI-generated documents, marking a significant development in how courts view the intersection of artificial intelligence and legal professional privilege. While the rules around privilege are slightly different in the United States, the concepts are broadly similar, and so this ruling should serve as a wake-up call, not only for American practitioners but also for Australian lawyers and their clients.

In February 2026, the U.S. District Court for the Southern District of New York held that AI‑generated documents prepared by a defendant and later shared with his lawyers were not protected by attorney‑client privilege or the work product doctrine.

The facts are captivating: Bradley Heppner was once the polished founder of a major U.S. financial services company, but by late 2025 he knew federal investigators were circling. Allegations of securities fraud, wire fraud and the misappropriation of more than USD 300 million were already on the government’s radar, and he had even received a grand jury subpoena. As the pressure mounted, Heppner turned not to his lawyers but to a public generative AI tool, Claude, feeding it sensitive information including details learned in privileged conversations in an effort to shape defence strategies, draft arguments and make sense of the investigation closing in around him. He produced 31 AI generated “reports,” believing they would remain private reflections for later discussion with his legal team. But on 4 November 2025, when federal agents raided his home and seized his devices, they found every one of those AI created documents stored and retrievable. In court, his lawyers argued these were no more than thought notes prepared for the purpose of obtaining legal advice, but their position quickly collapsed. Counsel conceded they had never instructed Heppner to use AI, had not supervised the creation of the documents and only became aware of them after the fact. Judge Rakoff reacted with immediate scepticism, noting that Claude was “plainly not an attorney” and that its terms expressly disclaimed confidentiality or any attorney client relationship, and permitted the operators of Clause to use and disclose user inputs and outputs. The court ordered all 31 documents to be provided to prosecutors, transforming what Heppner believed was a private digital notebook into a detailed roadmap of his fears, theories and strategies, delivered straight into the hands of the authorities he was trying to defend himself against.

The Australian perspective

Whilst the recent US decision is not binding on Australian courts, our judiciary has historically drawn upon American precedents when addressing novel legal issues, particularly in areas involving technology. Australian courts have also shown willingness to strictly interpret privilege requirements.

Generally speaking, any ‘legal advice’ generated by an automated AI system is unlikely to ever attract advice privilege in Australia, as advice privilege can only attach to confidential communications between a person and a qualified legal practitioner, or to confidential processes directed by a lawyer, for the dominant purpose of giving or receiving legal advice or for use in current or anticipated litigation. An AI system is not a qualified legal practitioner.

There may be circumstances where the output of AI could attract litigation privilege, where the relevant exchange was for the dominant purpose or use in existing or reasonably anticipated litigation. However, this would typically require that:

1. the AI tool utilised by a person is not publicly available (i.e., a private, closed data-set tool);
2. lawyers have been engaged before the relevant document is produced, and those lawyers have commissioned the preparation of that document for litigation.

For example, a chronology drafted with the assistance of AI by uploading relevant documents and communications into the AI platform is unlikely to attract legal professional privilege if it is prepared by a client using a publicly available AI tool prior to engaging lawyers (as it was obtained independently outside of the lawyer-client relationship). However, the same type of document would be protected by privilege if the relevant documents and communications were provided by a person to their lawyer in the first instance, and then those documents were uploaded by the lawyer to the law firm’s proprietary AI tool for the purpose of preparing a chronology that would allow the lawyer to assess a potential litigation claim. We accordingly believe that, even though Mr Heppner plainly anticipated he may be the subject of legal proceedings, his forays with AI prior to engaging legal representation would similarly be unlikely to be protected by privilege in Australia.

Further, Mr Heppner’s use of ‘public’ AI system is likely to have jeopardised the continued protection of those documents.  Most AI platforms retain user inputs for training purposes or quality improvement. When you feed confidential information into an AI system, you may be disclosing that information to third parties—the AI provider, its contractors, and potentially anyone who later interacts with the trained model.

Courts are typically unwilling to continue to provide privileged status to documents where the client has not handled those materials in a way which is consistent with the protection sought.  Failure to keep documents confidential can accordingly be fatal to privilege claims, which means that disclosure of privileged information to less secure AI systems could negate privilege.  An appropriately private AI system might not share those same issues.

Practical Guidance for Clients

The message is clear: seek legal advice first, use AI in connection with legal matters only under your lawyer’s guidance, and don’t feed copies of advice you receive from your lawyer to AI systems without understanding the associated risks.

Here’s what clients should do:

• Contact your solicitor at the outset. Before you draft documents, analyse contracts, or prepare legal communications, engage your lawyer. This ensures privilege attaches from the beginning.
• Don’t feed confidential information into public AI tools. Commercial AI platforms are not secure repositories for privileged information. What you share may be stored, analysed, or even disclosed. Don’t give ChatGPT a copy of the advice you receive from your lawyer and ask for a second opinion.
• Consult with your lawyer to help decide if and how to use AI. Many law firms now use AI tools within controlled, secure environments. Your solicitor can leverage these technologies whilst maintaining privilege protections.
• Disclose any prior AI use immediately. If you’ve already used AI tools before engaging your lawyer, disclose this immediately. Your solicitor needs to assess potential privilege implications and advise accordingly.
• Document your intent. When communications occur, make clear that you’re seeking legal advice for the dominant purpose of understanding your legal position or obtaining professional guidance.

Looking Ahead

The intersection of AI and legal privilege is still developing, and Australian courts will eventually face these questions directly. Until our courts provide definitive guidance, prudence demands that we treat AI-generated content with extreme caution when confidentiality and privilege are at stake.

The convenience of AI is undeniable, but the cost of waiving privilege – exposing confidential strategies, admissions, or advice to opposing parties – is simply too high. In legal matters, privilege is not merely a technical rule; it’s a cornerstone of effective representation and access to justice.

Australian clients should proceed on the assumption that AI is not your lawyer, and anything disclosed to a public platform may ultimately be used as evidence.

When facing a legal issue, your first conversation should be with your lawyer, not with an AI. Protect your privilege, protect your position, and ensure that your confidential communications remain exactly that – confidential.

This article was written by Daniel Kiley, Partner, Guzyal Hill, Special Counsel and Caitlin Surman, Special Counsel.