Are Discussions with AI Privileged? What United States v. Heppner Means for Your Case

Marguerite Seymour, Of Counsel

Mark Healy, Of Counsel

 

A recent US decision indicates that information a client enters into a consumer AI chatbot about an ongoing case is unlikely to be protected by legal professional (or lawyer-client) privilege. This may remain the position even where the client intended the material to assist their defence and later sends the AI-generated documents to their legal team. Mark Healy and Marguerite Seymour explain the decision and how clients should, and should not, use AI tools while a legal matter is ongoing. 

Key takeaway: Do not enter case facts, legal advice or case strategy into a consumer AI tool without first speaking to your legal team, as doing so may expose the material and may waive privilege in the underlying lawyer-client communications. 

 

What is lawyer-client privilege? 

Lawyer-client privilege protects confidential communications between clients and their lawyers that are made for the purpose of obtaining or giving legal advice. It is a bedrock of the legal system because it allows clients to speak openly and honestly with their legal representatives. With full information about the strengths and weaknesses of a case, lawyers can provide the best possible advice and strategy while protecting their clients’ interests. 

 

What did the Court decide in United States v Heppner? 

On 10 February 2026, Judge Rakoff of the US District Court for the Southern District of New York1 ruled that documents a criminal defendant had generated using a consumer version of an AI chatbot were not protected by lawyer-client privilege. On 17 February 2026, the judge issued a written opinion confirming and expanding on that reasoning. This appears to be the first reported decision squarely addressing whether a person’s private exchanges with a commercial AI tool, later shared with a legal team, can be protected from disclosure by lawyer-client privilege. 

The defendant, a corporate executive facing securities and wire fraud charges, used a free consumer-grade AI chatbot to prepare roughly 30 documents analysing the government’s case against him after he had received a grand jury subpoena and retained counsel. Some of the information he entered into the AI tool included advice his lawyers had given him. His lawyers had not instructed him to use AI; he did so entirely of his own volition. 

The defendant later sent the documents to his legal team. When they were seized during a search of his home, his lawyers asserted that the documents were protected by lawyer-client privilege. The government applied to compel their production, and the Court agreed that they should be produced because they were not privileged. 

 

Why did lawyer-client privilege not apply? 

The US Court found that privilege did not apply for three principal reasons: 

• The exchanges did not involve communication with a lawyer. The Court held that an AI platform is not a lawyer and cannot form a lawyer-client relationship, regardless of how the tool is marketed or used. The government compared the exchange to a client asking a friend for input: potentially informative, but not privileged. 

• There was no reasonable expectation of confidentiality. The Court referred to the AI provider’s privacy policy, which permitted prompts and outputs to be logged, used to train the model and disclosed to regulators or other third parties. Sending information to an AI tool operating under those terms was inconsistent with an expectation of confidentiality, regardless of what later happened to the documents. 

• The process was not undertaken to obtain legal advice. Although defence counsel argued that the client had used the tool for the purpose of talking to his lawyers, the Court did not accept this because counsel had not directed him to use it. The AI system itself stated that it could not provide legal advice and recommended consulting a qualified lawyer. Later forwarding the documents to counsel did not retrospectively make them privileged. 

 

Can using AI waive privilege in the original legal advice? 

Perhaps the most consequential aspect of the ruling was not the status of the AI-generated documents themselves, but the effect of entering privileged information into the tool. The government argued, and the Court accepted, that feeding advice received from lawyers into a non-confidential third-party platform can waive privilege over the underlying lawyer-client communications. In other words, the risk is not limited to the AI transcript being discoverable: the original conversation with counsel may also lose its protection. 

The Court noted that the outcome might differ where counsel directs a client to use a specific AI tool as part of the representation. In that situation, the AI platform might potentially function as counsel’s agent, in a manner comparable to expert accountants or investigators retained to assist with legal advice. The Court also suggested, without deciding, that an enterprise AI tool with contractual confidentiality commitments and no data-training rights might be analysed differently from a free consumer product. 

Neither point was resolved. Clients should therefore not assume that switching to a paid or enterprise product will cure the privilege problem. 

 

How are Irish and other courts approaching legal use of AI? 

Across common law jurisdictions, courts have increasingly warned about the risks of using AI in litigation, particularly the citation of hallucinated authorities. Irish courts have been vocal critics. Nolan J and Simons J have cautioned against AI’s unreliability, with Simons J noting that an applicant’s submissions appeared to have been AI-generated. 

In Guerin v Doherty2, Costello P set out five guiding principles after a party’s AI-hallucinated submissions caused wasted costs: 

• Parties may use AI for research but must not mislead the Court with unfounded propositions. 

• AI use should be disclosed to the Court and the opposing party. 

• Self-represented litigants bear the same responsibility for their submissions as lawyers do for theirs. 

• Parties must independently verify all AI-assisted research. 

• No authority should be cited without confirming that it is genuine and supports the proposition for which it is relied on. 

Other jurisdictions have expressed similar concerns. English courts in Bandla v Solicitors Regulation Authority3 and Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank QPSC and QNB Capital LLC4 have struck out cases and stressed lawyers’ duty to verify AI outputs independently. New Zealand’s Supreme Court in Jones v Family Court at Whangārei5 warned that reliance on false AI citations may amount to contempt of court. The Canadian courts have gone further by imposing personal costs consequences on lawyers who rely on AI-hallucinated authorities. In Zhang v Chen6, the Supreme Court of British Columbia ordered costs against counsel personally after fictitious ChatGPT-generated cases were cited in family law proceedings and directed a review of the lawyer's other pending files. 

In May 2026, in the English case of Cork and Anor v Smith7, Pinsent Masons self-reported AI-related hallucinations to its regulator. In that case, Mullen J described the firm’s conduct as “deeply troubling”, reinforcing that AI cannot replace genuine legal research. 

 

What does Practice Direction HC 142 require? 

The President of the High Court, Mr Justice David Barniville, has issued Practice Direction HC 142, which comes into effect on 1 September 2026. It provides guidance on the Court’s expectations regarding the appropriate and responsible use of generative artificial intelligence (GenAI) tools in connection with legal proceedings. 

The Practice Direction applies to GenAI-related documents generated after 1 September 2026 and used in civil proceedings before the High Court. It applies to all persons involved in a case, including legal practitioners, witnesses, experts and any third party required by the Court to produce documents. 

Any person responsible for producing a GenAI document must ensure that it has been appropriately and independently verified before use in proceedings. Any limitation on their ability to do so must be disclosed to the Court. 

A person relying on a GenAI document must ensure that it is fact-checked and proofread; adapted to the particular circumstances of the case; and that references to case law, legislation, textbooks or articles exist and support the legal propositions attributed to them. Extracts and quotations must be verified as accurate and correctly attributed to the original source. 

Failure to comply can result in adverse orders, the Court disregarding or rejecting documents in whole or in part, or the striking out of a claim or defence where appropriate. A legal practitioner who fails to comply may be referred to the Legal Services Regulatory Authority and may face a personal costs order. 

 

What should clients do before using AI during a legal matter? 

Until there is further clarity from the courts, clients should follow these practical steps: 

• Do not discuss case facts, strategy or anything your solicitors or counsel have told you with an AI chatbot without first seeking advice. Do not use an AI tool to create legal documents where doing so relies on information provided by your legal team. 

• If an AI tool could genuinely assist your matter, speak to your legal team before using it. They can advise whether and how a specific tool might be used without jeopardising privilege, including whether it should be used at all. 

• Assume that anything already entered into a consumer AI tool about your matter may be discoverable. If you have concerns, raise them with your legal team promptly. 

• Do not assume that a paid subscription or enterprise label automatically solves the problem. The protections depend on the specific terms of service and how the tool is used. 

• Ensure that employees, family members and colleagues assisting you follow the same guidance. Privilege can be lost through their use of AI just as easily as through your own. 

 

What are the next steps? 

Holmes is monitoring developments in this area closely, including how courts distinguish between consumer and enterprise AI tools and approach AI in the context of legal professional privilege. Before using any AI platform in connection with an ongoing or anticipated legal matter, contact our Litigation & Dispute Resolution team, Mark Healy or Marguerite Seymour for advice on your specific circumstances. 

 

Dislaimer of Liability: This insight article is provided for general information purposes only and should not be relied upon as legal advice. Readers should seek specific legal advice tailored to their individual circumstances before taking or refraining from any action. Holmes O'Malley Sexton LLP, its partners, employees, agents, and affiliated entities disclaim all liability, to the fullest extent permitted by law, for any loss or damage arising directly or indirectly from reliance on the information contained in this publication.