In February 2026, a federal court in New York held in United States v. Heppner (Heppner) that a litigant’s communications with a public generative artificial intelligence (AI) platform were neither privileged nor protected work product. In the months since, the question of whether AI use can waive privilege or work product protection has been litigated several times across the United States, with varying outcomes.
This post is a follow-up to our April 2026 blog, “When AI meets privilege: What foreign courts are telling us about the risks of public AI in legal practice”.[1] We summarize three American decisions that have considered Heppner, then discuss how Canadian law, still untested on this question, may respond.
Developments in American case law since Heppner
In United States v. Heppner, Judge Rakoff held that communications between a criminal defendant and a consumer AI platform were not protected by attorney-client privilege or the work product doctrine: the platform was not a lawyer, its terms of use defeated any reasonable expectation of confidentiality, and the materials were not prepared at counsel’s direction.[2] In the United States, the work product doctrine protects documents and materials prepared in anticipation of litigation by or for a party or its representative, and is broadly comparable to litigation privilege in Canada. Subsequent US case law has not consistently followed Heppner. Two approaches have emerged. Some courts, like Heppner, have denied protection where AI was used, while others have held that AI-assisted litigation preparation can be protected work product, and that using an AI tool does not by itself waive that protection because the tool is not an adversary.
Client AI use without counsel
In a dispute over the sale of a Boston gas station, the plaintiff forwarded a default letter to his romantic partner, who used ChatGPT to generate a response letter and other documents without any lawyer’s involvement.[3] The defendants moved to compel production, and the plaintiff argued the materials were protected work product. The Massachusetts Superior Court ordered the plaintiff to produce the ChatGPT communications and output. It held that the partner was not a party’s “representative” for work product purposes, and that neither AI queries for legal assistance by such a person nor the resulting output were protected unless performed at the direction of counsel.[4] The court found the facts similar to Heppner, where the AI documents were not prepared at counsel’s behest and did not reflect counsel’s strategy.[5] It also held that ChatGPT’s output is not opinion work product: “AI is a tool, not a person.”[6]
Work product for the self-represented, with limits
In an employment discrimination case, the defendant sought to amend the protective order to address AI and to compel the self-represented plaintiff to identify the AI tool he was using. The plaintiff refused, arguing that his choice of tool was protected work product. The court held that the federal work product rule protects a self-represented litigant’s AI-assisted litigation preparation, distinguishing Heppner because a self-represented litigant is both party and advocate. However, the plaintiff was ordered to name any AI tool he had used with confidential information, and the court barred inputting confidential information into any AI platform unless the provider is contractually prohibited from training on or disclosing inputs and must allow deletion on request, a condition the court acknowledged would, for now, exclude most mainstream low-to-no-cost AI tools.
The court also considered waiver of privilege. The issue was whether entering information into a public AI tool, whose provider collects user data, amounted to waiver. The court reasoned that routing information through a third-party system does not forfeit all privacy, and that work product is typically waived only by disclosure to an adversary or in circumstances that substantially increase that likelihood. AI interactions therefore do not automatically compromise protection.
In Tate Group, the plaintiff claimed work product privilege over ChatGPT conversations generated by its principal. The Texas Business Court reviewed the conversations in camera and rejected the defendants’ arguments that work product does not extend to a non-lawyer’s AI chats and that AI use waived any protection.[7] Adopting Morgan’s waiver standard, the court expressly stated that it “disagrees with” Heppner, noting that Texas Rule of Civil Procedure 192.5(a)(1), unlike the federal rule, protects “material prepared or mental impressions developed in anticipation of litigation or for trial by or for a party.”[8] Most of the conversations could be withheld, though certain pages were ordered produced and, following Morgan, the plaintiff was ordered to identify all discovery materials it had shared with ChatGPT.[9]
Taken together, these decisions suggest an emerging distinction among courts in the United States. Courts have been more willing to protect AI materials as work product where the user is a party preparing its own case, and less willing where a represented client, or someone acting for them, uses AI independently of counsel. Even where protection is recognized, courts are requiring disclosure of the tools used and restricting what confidential information may be input into public platforms.[10]
Canadian privilege law and AI
Canadian case law on AI and privilege is still developing. To date, there has been no foundational Canadian decision addressing whether inputting confidential or privileged information into a generative AI tool waives solicitor-client or litigation privilege, or whether AI-generated material can itself attract privilege.[11]
Litigation privilege protects documents and communications whose dominant purpose is preparation for pending or reasonably apprehended litigation, creating a “zone of privacy” for the adversarial process.[12] Unlike solicitor-client privilege, it is not directed at lawyer-client communications as such: it extends to communications with third parties and to non-confidential material, applies to unrepresented litigants, and generally ends with the litigation.[13] Because confidentiality is not essential to it, litigation privilege may prove more resilient to AI use. A key question will likely be when the AI material was created and why, rather than who else may have had access to it.
Solicitor-client privilege protects a communication between lawyer and client that entails the seeking or giving of legal advice and is intended to be confidential.[14] It protects a relationship rather than a process, and it does not end when the matter or litigation concludes: once established, it continues indefinitely unless waived by the client.[15] Because it depends on confidentiality, it is more vulnerable to AI. Canadian courts have not yet decided whether a public AI platform can stand in the place of a lawyer, but if they follow the US approach in Heppner, it will not. Inputting privileged communications into a tool whose terms permit the provider to store, review or train on that data may then be characterized as a disclosure to a third party that is inconsistent with confidentiality, and therefore a waiver.[16]
BC case law
No court in British Columbia, or elsewhere in Canada, has addressed whether AI use affects privilege, but existing BC case law on third parties offers some indication of how that question may be answered. In College of Physicians and Surgeons of B.C. v. British Columbia (Information and Privacy Commissioner), the BC Court of Appeal held that third-party communications are protected only where the third party acts as a channel of communication or performs a function, on the client’s behalf, “integral to the relationship between the solicitor and the client,” as opposed to merely gathering information or providing opinions.[17]
Applying that framework, the BC Supreme Court in Starrs v. Troczynski held that a third-party investigative report was not subject to solicitor-client privilege merely because counsel retained the investigator and used the report for legal advice; gathering information and passing it to counsel is insufficient.[18] By analogy, AI used to gather, summarize or analyze information for counsel may struggle to fall within solicitor-client privilege, unless it can be characterized as a conduit, akin to a translator, for lawyer-client communications.
The analysis may be more favourable under litigation privilege. In Lo v. Penticton (City), the court held that litigation privilege applies where litigation was reasonably in prospect and was the document’s dominant purpose, and that distinct litigation-related portions may be severed and protected.[19] Applied to AI, litigation privilege may protect a prompt, transcript or AI-assisted draft where: (a) litigation was reasonably in prospect when the material was created; (b) its dominant purpose was obtaining legal advice or preparing for litigation; and (c) the claim is established for the particular record or output, rather than the entire AI workspace.[20] That approach would broadly align with the American decisions that have protected AI materials prepared for litigation as work product.
Takeaways
US courts are divided. Heppner and Shealy show that AI use by a represented client acting without counsel can forfeit protection, while Morgan and Tate Group show that AI-assisted litigation preparation may still be protected work product.[21]
In Canada, solicitor-client privilege is likely the more vulnerable of the two privileges, because it depends on confidentiality and on any third party performing a function integral to the solicitor-client relationship. Litigation privilege, which turns on timing and dominant purpose, may offer stronger protection for AI-assisted litigation work.
The use of AI in connection with a legal matter may carry real risks to privilege. Uploading privileged material to a public platform may be treated as a disclosure to a third party, and AI use undertaken by a client without counsel’s involvement may not be protected.
For more information on this topic, reach out to the authors, Morgan Camley, KC, Liam Edmondstone and Kathryn Gullason.
[1] Morgan Camley, KC, Kay Scorer & Liam Edmondstone, “When AI meets privilege: What foreign courts are telling us about the risks of public AI in legal practice” (28 April 2026), online: Dentons Commercial Litigation Blog <www.commerciallitigationblog.com/when-ai-meets-privilege-what-foreign-courts-are-telling-us-about-the-risks-of-public-ai-in-legal-practice/> [Camley, Scorer & Edmondstone].
[2] United States v Heppner, No 25-CR-503 (JSR), 2026 WL 436479 (SDNY 2026) at *3–4 [Heppner]; see also Camley, Scorer & Edmondstone, supra.
[3] Earl Wade Shealy Jr v Seaside Investments, LLC, No 2684CV00799-BLS2 (Mass Super Ct 16 June 2026) [Shealy], as summarized in Matthew W Costello & Jack J Tierney, “Lessons from Shealy v. Seaside: When client AI use can sink work product protection” (29 July 2026), online: Nixon Peabody LLP <www.nixonpeabody.com/insights/alerts/2026/07/29/lessons-from-shealy-v-seaside-when-client-ai-use-can-sink-work-product-protection> [Costello & Tierney].
[4] Costello & Tierney, supra.
[5] Ibid.
[6] Ibid.
[7] Tate Group Automotive, LLC v Legacy Automotive Capital, LLC, Cause No 25-BC11B-0020 (Tex Bus Ct 3 June 2026) (minute entry) at 1–2 [Tate Group].
[8] Tate Group, supra at 2–3, citing Tex R Civ P 192.5(a)(1).
[9] Tate Group, supra at 3–4.
[10] See Costello & Tierney, supra; Morgan, supra at 11–15; Tate Group, supra at 3–4.
[11] Jennifer E Brevorka & Tanzim Rashid, “Technology, Third Parties, and Maintaining Privilege” (2025) 55:3 Adv Q 399 at 417 [Brevorka & Rashid]; Camley, Scorer & Edmondstone, supra.
[12] Blank v Canada (Minister of Justice), 2006 SCC 39 at paras 27, 59–60 [Blank].
[13] Lizotte v Aviva Insurance Company of Canada, 2016 SCC 52; Blank, supra at paras 27, 33–39; Brevorka & Rashid, supra at 402.
[14] Solosky v The Queen, [1980] 1 SCR 821; Brevorka & Rashid, supra at 401.
[15] Lizotte v Aviva Insurance Company of Canada, supra.
[16] Brevorka & Rashid, supra at 425; Camley, Scorer & Edmondstone, supra.
[17] College of Physicians and Surgeons of BC v British Columbia (Information and Privacy Commissioner), 2002 BCCA 665 at paras 47–50 [College of Physicians].
[18] Starrs v Troczynski, 2024 BCSC 2267 at paras 22–27 [Starrs].
[19] Lo v Penticton (City), 2022 BCSC 2230 at paras 20–31, 47–52 [Lo].
[20] See Lo, supra at paras 22–27, 31, 52; College of Physicians, supra at para 72.
[21] Costello & Tierney, supra; Tate Group, supra at 2.