There’s no ‘AI’ in Attorney! The cautionary tale of generating legal documents through artificial intelligence and the strict rules around legal professional privilege, as told in United States v Heppner.
The use of AI amongst Australians has increased dramatically over the last few years. Writing a shopping list, planning a holiday, or sending a breakup text can now be done in a matter of seconds on the palm-sized-supercomputer that sits in your pocket. But what happens when you start consulting your loyal AI friend to develop legal strategy for your impending indictment, then try to withhold those documents from the court later on? The claim for legal professional privilege (or attorney-client privilege) over AI generated documents was tested recently in the case of United States v Heppner, as precedents regarding AI begin to develop around the world.
What happened in United States v Heppner (Heppner)?
In October 2025, Bradley Heppner was charged with securities fraud, wire fraud, conspiracy to commit securities fraud and wire fraud, making false statements to auditors, and falsifying corporate records. The indictment charges concerned Heppner’s alleged misconduct as an executive of several corporate entities including the publicly traded company GWG Holdings, Inc.
Mr Heppner pleaded not guilty to these charges on 10 November 2025.
After discovering he was the target of a criminal investigation, Mr Heppner used a well-known generative AI platform called ‘Claude’ (powered by Anthropic) to “prepare reports that outlined defence strategy, that outlined what he might argue with respect to the facts and the law that we anticipated that the government might be charging.”[1] He did so on his own accord, without the direction of counsel. A total of 31 documents were created with the assistance of Claude. The prompts and AI‑generated content was later shared with Mr Heppner’s lawyers.
Mr Heppner’s lawyers argued that he was “was preparing these reports in anticipation of a potential indictment”[2]
Mr Heppner and his lawyers tried to assert privilege over the 31 AI generated documents, arguing that:
- Mr Heppner had input information that he had learned from counsel into Claude;
- that he had created the AI documents with the purpose of speaking about them with counsel to obtain legal advice; and
- that he subsequently shared the documents with counsel.
Why the claim for privilege failed
The Court determined that Mr Heppner’s claim for attorney-client privilege over the documents failed, and cited three reasons as to why:
Firstly, the communications were not between Mr Heppner and counsel. One of the quintessential elements of claiming privilege is the existence of a trusting human relationship between a client and a licenced professional, who “owes fiduciary duties and is subject to discipline”[3]. The mere fact that Claude is not a lawyer is enough to dispose of a claim for attorney client privilege, in the same way that any conversation between any two non-lawyers would be. Confiding in your friend at the pub about an upcoming legal matter would be treated the same way!
Secondly, the communications memorialised in the AI documents were in no way confidential. By submitting information to almost any AI platform, you have shared it with a third party. Claude’s written privacy policy, which all users must consent to, provides that Anthropic collects data on both users’ “inputs” and Claude’s “outputs,” that it uses such data to “train” Claude, and that Anthropic reserves the right to disclose such data to a host of “third parties,” including “governmental regulatory authorities”[4]. Further, Claude’s policy states that Anthropic may “disclose personal data to third parties in connection with claims, disputes [,] or litigation,” even in the absence of a subpoena demanding the company do so.
Finally, it was determined that Mr Heppner did not communicate with Claude for the purpose of obtaining legal advice. In fact, during the proceedings, the prosecution asked Claude if it could give legal advice, and it replied that “I am not a lawyer and can’t provide legal advice or recommendations” and recommended that the prompter should “consult with a qualified attorney who can properly assess your specific circumstances”. [5] Mr Heppner’s counsel argued that Mr Heppner communicated with Claude for the express purpose of later talking to counsel but also conceded that Heppner did not make these communications at counsel’s direction, but rather of his own volition. Judge Rakoff makes a comment in his memorandum that had Mr Heppner’s counsel directed him to use Claude, there may have been an argument for Claude to be classified as a sort of ‘attorney’s agent’, whose communications could be covered within the remit of attorney-client privilege. The relevant question here is whether Mr Heppner intended to obtain legal advice from Claude, not whether he shared Claude’s outputs with his lawyers.
Judge Rakoff further noted in his memorandum that the communications were not privileged at the time they took place. Black letter law establishes that just because you share non-privileged communications with counsel does not somehow alchemically make them privileged. Material does not acquire special protection just because it is handed to a professional whose work may be subject to that protection.
ChatGPT Privacy Policy
ChatGPT (powered by OpenAI) is certainly the most used free AI chatbot in Australia. It is safe to assume that similar judgements to the one discussed above would be made in Australian courts, in relation to trying to assert privilege over conversations with ChatGPT, Co-Pilot, Grok or any other artificial intelligence, when seeking legal advice without the direction of a lawyer.
ChatGPT’s privacy policy states that OpenAI collects and processes “User Content”, which includes prompts, documents, images, audio and other inputs you provide when using the service. The policy also allows for information to be stored, reviewed, analysed or reused internally by the bot. OpenAI may also share personal data with affiliates and contractors, and disclose it to service providers who assist with hosting, infrastructure, security or analytics. Further, the policy confirms that data may be transferred outside your jurisdiction (including to the United States).
Any information you input into ChatGPT is legally and technically shared with OpenAI as a third party, may be stored and used in accordance with its privacy policy, and will not be treated as confidential or privileged information by a court.
What does legal professional privilege mean in Australia?
Legal professional privilege is a common law protection that allows a person to resist demands to disclose information or produce documents which would reveal communications made between a client and their lawyer, which were made for the dominant purpose of giving or obtaining legal advice.
Legal professional privilege ‘exists to serve the public interest in the administration of justice by encouraging full and frank disclosure by clients to their lawyers’[6]. The High Court has confirmed that privilege is not a mere rule of evidence but a fundamental common law right.[7]
What you need to know
Be cautious of using AI tools for anything regarding upcoming litigation or legal issues, as the privacy policy and data handling terms of that AI bot could seriously undermine a retrospective claim for privilege over any prompts or outputs.
Consult lawyers first. Providing us with documents you have prepared by AI risks our ability to claim privilege over them later.
Remember, just because you have a lawyer involved in your matter does not entitle all documents to privilege. As a non-lawyer, using AI to determine legal risk or formulate defences, without the direction of a lawyer and without the predominant purpose being obtaining legal advice, anything produced will likely fall outside the ambit of professional privilege, and therefore be subject to scrutiny of the Court or the other side.
Use Lavan (or at least another lawyer)!
We understand that AI has seeped into many facets of life, including everything from personal organisation to streamlining large business operations. Google presents an “AI Overview” before presenting any search results, Grammarly finishes your sentences for you, and it seems like we are being introduced to a new AI assistant on a new platform every day; from Grok on X (the app formally known as Twitter), to Tako on TikTok, to Meta AI on Facebook. With such a focus on maximising efficiency in every part of our lives, it is not surprising that clients may consider turning to AI to help them when facing a legal battle.
However, the case of Heppner has given us a glimpse into how courts around the world may begin to treat communications with AI in legal settings. The way ‘privilege’ relating to AI-produced documents and communications is handled will continue to develop, and it is not farfetched for us to say that an Australian case concerning this issue is surely on the horizon. Heppner has also sparked much commentary in the legal community. Particularly about why in this matter privilege was denied, when previous matters, documents prepared by clients in Google Docs, Google Slides, Gmail, and stored in iCloud, (i.e., other third-party platforms) courts seemingly do not ask whether the use of those intermediary tools defeats claims of privilege.[8]
This is an area that is sure to evolve, and fast. Keep an eye on Lavan’s Cyber Update Publications for future developments. And, if you are having trouble developing your legal arguments yourself, don’t turn to ChatGPT – contact the Dispute Resolutions and Investigations team at Lavan for privilege-protected advice instead!
Many thanks to solicitor Isabella Beilin for her assistance in the preparation of this publication.
Disclaimer
The information contained in this publication does not constitute legal advice and should not be relied upon as such. You should seek legal advice in relation to any particular matter you may have before relying or acting on this information. The Lavan team are here to assist.
Footnotes
[1] Memorandum 25 Cr. 503 (JSR) of Jed D Rakoff U.S.D.J, page 3
[2] Memorandum 25 Cr. 503 (JSR) of Jed D Rakoff U.S.D.J, page 3
[3] United States v Heppner, Memorandum, United States District Court for the Southern District of New York, Rakoff J, Case No 25 Cr 503 (JSR), 17 February 2026, page 6.
[4] United States v Heppner, Memorandum, United States District Court for the Southern District of New York, Rakoff J, Case No 25 Cr 503 (JSR), 17 February 2026, page 6.
[5] United States v Heppner, Memorandum, United States District Court for the Southern District of New York, Rakoff J, Case No 25 Cr 503 (JSR), 17 February 2026, page 8.
[6] Esso Australia Resources v Commissioner of Taxation (1999) 201 CLR 49, [35] (Gleeson CJ, Gaudron and Gummow JJ).
[7] Esso Australia Resources Ltd v Commissioner of Taxation (1999) 201 CLR 49.
[8] As detailed in the Harvard Law Review: United States v. Heppner – Harvard Law Review
Related Publications
Australia’s cyber security strategy 2023 to 2030 – progress to date and the shift to Horizon 2
Australian Information Commissioner v Australian Clinical Labs (No 2) [2025] FCA 1224 – First civil penalty ordered in the history of the Privacy Act 1988 (Cth)
Stay up to date with Lavan
"*" indicates required fields