The legal profession’s relationship with artificial intelligence is evolving rapidly, and not always smoothly. Canadian courts and tribunals are grappling with the consequences of AI misuse in litigation, from fabricated case citations to undeclared reliance on generative tools. A more fundamental question is now emerging from the United States, the United Kingdom and Australia: can the use of public AI tools compromise solicitor-client privilege?
AI in Canadian courts
Decisions addressing AI misuse now span the full spectrum of Canadian courts and tribunals, from the Civil Resolution Tribunal in British Columbia to Courts of Appeal.
The cases share common themes. In Wu v. Murray, the BC Court of Appeal Registrar held a self-represented appellant accountable for AI-generated false cases, citing the Court’s filing directive that reminds “all litigants that they are responsible for the authenticity and accuracy of all materials filed with the Court.”[1] In Simpson v. Hung Long, a claimant’s submissions cited fabricated Supreme Court of Canada decisions written in what the tribunal described as a “convincingly legal tone,” leading to the claim’s dismissal and a CA$500 compensation order for the opposing party’s time.[2]
Perhaps most consequentially, in Reddy v. Saroya, the Alberta Court of Appeal ordered appellant’s lead counsel to personally pay CA$17,550 in costs after a factum drafted by a contractor using an AI ‘large language model’ contained references to seven non-existent case authorities, representing nearly half of the cases in the table of authorities.[3] Large language models are a type of artificial intelligence system trained on massive volumes of text data to understand and generate human language, which in the legal context can be used for tasks such as drafting documents and conducting legal research, but carries a well-documented risk of producing fabricated case citations or “hallucinations.”[4] The Court applied Rule 10.50 of the Alberta Rules of Court, finding the conduct constituted a “marked departure from reasonable conduct” that seriously interfered with the administration of justice.
Canadian courts take AI misuse seriously at every level. These assessments, however, have primarily addressed the problem of fabricated authorities and the duty not to mislead the court. The next frontier, now being explored abroad, is whether the use of public AI tools has the potential to undermine solicitor-client privilege itself.
Solicitor-client privilege
Solicitor-client privilege is one of the most important protections in Canadian law, having been described by the Supreme Court of Canada as “a principle of fundamental justice and civil right of supreme importance in Canadian law.”[5] It shields confidential communications between a lawyer and their client, made for the purpose of seeking or providing legal advice, from compelled disclosure. The Supreme Court of Canada has described it as a principle of fundamental justice, essential to the proper functioning of the legal system.[6]
The privilege rests on two pillars: confidentiality and the lawyer-client relationship. A communication must be intended to be confidential, and that confidentiality must be maintained. If a client voluntarily discloses the substance of privileged communications to a third party, the privilege may be waived.[7]
The overseas spillover: AI and privilege beyond Canada’s borders
Solicitor-client privilege is a feature of the legal systems of other major common law jurisdictions. In the United States, the attorney-client privilege is an evidentiary privilege derived from the common law that protects confidential communications between a client and their attorney made for the purpose of seeking or obtaining legal advice.[8]
The United States also has a functional equivalent to Canada’s litigation privilege, called the “work product doctrine.” Under Canadian law, litigation privilege is a privilege that protects communications and documents created for the purpose of anticipated litigation.[9] Similarly, the work product doctrine provides qualified protection for materials prepared by counsel in anticipation of litigation, and its purpose is to preserve a zone of privacy in which a lawyer can prepare and develop legal theories and strategy.[10] In that respect, Canada’s litigation privilege and the United States work product doctrine are functionally analogous, because both doctrines are designed to provide a protected zone for counsel’s litigation preparation, including the development of legal theories and strategy in anticipation of litigation.
In the United Kingdom, the analogous doctrine to Canada’s solicitor-client privilege takes the form of “legal professional privilege” which itself divides into legal advice privilege, covering confidential communications between lawyer and client made for the dominant purpose of giving or receiving legal advice, and litigation privilege, which extends to confidential communications with third parties made for the dominant purpose of contemplated or existing adversarial proceedings.[11]
In Australia, legal professional privilege is both a common law right and a creature of statute. Australia’s Evidence Act [12] protects confidential communications and documents created for the “dominant purpose” of obtaining legal advice or in connection with actual or anticipated litigation, and it is regarded as a fundamental right vesting in the client. Despite minor variations in doctrine, all three jurisdictions ground the privilege in the same core principles: the communication must be confidential, it must arise within a lawyer-client relationship and its purpose must be the seeking or giving of legal advice.
Courts in other common law jurisdictions have begun to confront a question Canadian courts have not yet addressed: what happens to privilege when a client or lawyer feeds confidential information into a publicly available AI platform? Answers are emerging from the United States, the United Kingdom and Australia. Although each jurisdiction applies its own doctrinal framework, the underlying reasoning draws on the same orthodox principles of confidentiality, the purpose of the communication and the nature of the relationship that are fundamental to solicitor-client privilege in Canada.
United States v. Heppner: The AI platform is not your lawyer
The United States District Court for the Southern District of New York in United States v. Heppner,[13] decided on February 17, 2026, was directly confronted with the relationship (or lack thereof) between AI and privilege. Judge Jed S. Rakoff addressed what the court described as a question of first impression nationwide: whether communications between a user and a publicly available AI platform are protected by attorney-client privilege or the work product doctrine.
A detailed overview of the outcome of United States v. Heppner is available here. The analysis considers the court’s finding that a litigant’s communications with a publicly available generative AI tool were not protected by attorney-client privilege or the work product doctrine. It also examines the reasoning underpinning that finding, including the absence of a reasonable expectation of confidentiality under the platform’s terms of use, the lack of a lawyer-client relationship and the fact that the AI tool was not providing legal advice.
UK v. Secretary of State for the Home Department: Privilege waived by upload
The United Kingdom Upper Tribunal’s decision in UK v. Secretary of State for the Home Department,[14] handed down on November 17, 2025, consolidated two matters in which fictitious authorities had been submitted to the Tribunal following the use of AI-generated documents. The Tribunal used the occasion to issue sweeping guidance on the professional obligations of legal practitioners in the AI era.
The first matter involved Mr. Tahir Mehmood Mohammed of TMF Immigration Lawyers, who included a non-existent case in grounds of appeal. After initially denying AI use, Mr. Mohammed ultimately concluded that the false citation was probably the result of inadvertent use of an AI large language model, stating: “In absence of an explanation and with how AI operates, I cannot dismiss the fact that the case was an AI creation as there is no other explanation.”[15] The Tribunal’s investigation revealed that Mr. Mohammed had also been uploading client emails and Home Office decision letters into ChatGPT to generate summaries for clients.
It was this practice that prompted the Tribunal’s most significant pronouncement on privilege. The Panel stated in unequivocal terms: “to put client letters and decision letters from the Home Office into an open source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege.”[16]
The second matter involved Mr. Zubair Rasheed of City Law Practice Solicitors, who bore responsibility for the application form for judicial review drafted by a part-time trainee, Mr. Waheed Malik, that contained multiple fictitious authorities. The Tribunal’s investigation revealed serious supervisory failures, inconsistent explanations and inadequate record-keeping. The Tribunal observed that “it would be easy to think that this is a case about the naïve use of generative AI, but it is not merely about that; it is principally about supervision and the obligation to ensure that the tribunal is not misled.” The Tribunal held that “a supervisor who fails to ensure that the work of a more junior fee-earner does not contain false cases or citations is likely to be more culpable than a lawyer who fails to ensure that his own work is free from such ‘hallucinations’.”[17]
Helmold & Mariya (No 2): The Australian caution
Australia has not yet produced an authoritative ruling on AI and privilege, but the Federal Circuit and Family Court of Australia sounded a clear warning in Helmold & Mariya.[18] In that case, the appellant, a self-represented litigant, had used generative AI to prepare his appeal documents, which contained fictitious case citations.
In addressing the use of AI, the Court went beyond the immediate issue of fabricated authorities to flag the privilege implications of AI use in family law proceedings. The Court observed that “if a person inputs court documents into an open AI program, we consider that this may have the potential to fall foul of the provisions which prohibit communication of an account of proceedings to the public or a section of the public.”[19] The Court further stated that the “input of documents arising out of the proceedings into a generative AI program which stores, collates and replicates data may waive privilege or fall foul of the requirements that certain matters be treated as commercial in confidence.”[20] These issues, the Court concluded, warrant extreme caution.
What this means for Canadian practice
Canadian courts have yet to opine on whether the use of public AI tools can waive solicitor-client privilege. However, the reasoning in Heppner, the UK Tribunal’s decision and the Australian authorities all rest on principles that are familiar to Canadian law. Privilege depends on confidentiality. Confidentiality requires a reasonable expectation that the communication will not be disclosed. And voluntarily disclosing privileged information to a third party, whether that third party is a person, a corporation or a publicly accessible AI platform, is fundamentally inconsistent with maintaining that confidentiality.
If a Canadian court were to adopt reasoning similar to Heppner or the UK Tribunal’s pronouncement, the repercussions would be substantial. A finding that uploading privileged materials to a public AI tool constitutes waiver would expose clients and lawyers alike to serious consequences: previously shielded communications could become producible in litigation, regulatory investigations or criminal proceedings. Before using any AI tool, lawyers and clients must carefully consider whether doing so could compromise the confidentiality that privilege depends upon.
For more information on this topic, please reach out to Morgan Camley, KC, Kay Scorer and Liam Edmondstone.
[1] Section 7.3 of the BC Court of Appeal Registrar’s Filing Directive: “Given the rapid development of artificial intelligence tools, the Court reminds all litigants that they are responsible for the authenticity and accuracy of all materials filed with the Court.”; Wu v. Murray, 2025 BCCA 365
[2] Simpson v. Hung Long Enterprises Inc., 2025 BCCRT 525
[3] Reddy v. Saroya, 2026 ABCA 20
[4] Fatemeh Dehghani et al, “Large Language Models in Legal Systems: A Survey” (2025) 12 Humanities and Social Sciences Communications 1977, online: Nature https://doi.org/10.1057/s41599-025-05924-3
[5] In Lavallee, Rackel & Heintz v. Canada (Attorney General), 2002 SCC 61, para 36.
[6] Smith v. Jones, 1999 CanLII 674 (SCC), [1999] 1 SCR 455 at para 45
[7] Solosky v. The Queen, [1980] 1 SCR 821 at para 15, described solicitor-client privilege as: “(i) a communication between solicitor and client; (ii) which entails the seeking or giving of legal advice; and (iii) which is intended to be confidential by the parties”.
[8] Deborah R Meshulam et al. “Privilege Issues in Cross-Border Investigations Around the World” in The Guide to International Enforcement of the Securities Laws, 3rd ed (London: Global Investigations Review, 2023).
[9] Blank v Canada (Minister of Justice), 2006 SCC 39, [2006] 2 SCR 319.
[10] Jeff A Anderson et al, “Work Product Doctrine” (1983) 68:6 Cornell L Rev 760.
[11] Ashurst LLP, “Quickguide: Privilege under English Law” (2025), online: Ashurst <www.ashurst.com>; see also Three Rivers District Council v Governor and Company of the Bank of England (No 6), [2004] UKHL 48).
[12] Law Council of Australia, “Client Legal Privilege”(2016), online: Law Council of Australia <lawcouncil.au>; see also Baker v Campbell (1983), 153 CLR 52 (HCA)
[13] United States v Heppner, No 25-CR-503 (JSR), 2026 WL 436479 [“Heppner”]
[14] UK v. Secretary of State for the Home Department [2026] UKUT 81 (IAC) [“UK v. Secretary of State”]
[15] UK v. Secretary of State at para 13
[16] UK v. Secretary of State at para 21
[17] UK v. Secretary of State at para 38
[18] Helmold & Mariya (No 2), [2025] FedCFamC1A 163 [“Helmold”]
[19] Helmold at para 9
[20] Helmold at para 9
