The growing adoption of AI tools has extended into legal proceedings, particularly the use of generative AI to draft court materials (as an aside: the author expressly discloses that this blog was drafted with the aid of generative AI). However, AI-generated documents frequently contain “hallucinated” content—fabricated case law, fictitious statutes, and nonsensical arguments. As of February 23, 2026, an online database reported over 972 legal proceedings worldwide (61 in Canada) in which generative AI was used and produced hallucinated content.
When lawyers use generative AI, the risk does not arise from the mere use (privacy concerns aside), but from their failure to exercise due diligence in verifying accuracy. Such failures cause delay, introduce confusion, and may constitute an abuse of process that brings the administration of justice into disrepute. This concern is reflected in the reasons of the Alberta Court of Appeal in a recent case involving AI-generated submissions.
In Reddy v. Saroya, 2025 ABCA 322, dated September 26, 2025, the Alberta Court of Appeal considered an appeal from a chambers order finding the appellant in civil contempt for failing to provide adequate responses to undertakings in a commercial dispute. The appeal was allowed in part.
Of regulatory significance, however, was that the appellant’s lead counsel filed an AI-generated factum containing multiple fabricated authorities (para 74). The respondent had identified potential AI-related issues, but counsel advised that there were errors in the citations but the cases existed (para 74). Counsel later advised that a contractor had drafted the factum and had (incorrectly) assured him AI was not used (para 75). At the hearing, counsel acknowledged that, due to time constraints, he had failed to properly review the factum and detect the fictitious cases (para 79).
Addressing the AI-generated factum, the Court emphasized:
- The Law Society of Alberta’s (“LSA”) Code of Conduct requires lawyers to perform all legal services to the standard of a competent lawyer, including developing competence with relevant technology (para 80);
- The LSA has cautioned that, without safeguards, generative AI may “frequently introduce[s] confusion and delay into proceedings, and worse, constitute[s] an abuse of process that may potentially bring the administration of justice into disrepute” (para 80);
- An October 2023 Alberta Courts Notice requires meaningful human verification of all AI-generated submissions (para 82);
- Time required to verify and cross-reference AI-generated authorities must be incorporated into practice management responsibilities (para 83); and
- A lawyer who files a document whose name appears on the documents bears ultimate responsibility for its form and content, even where another individual prepared it (para 83).
The Court invited further submissions on whether costs should be imposed personally on counsel.
Subsequently in Reddy v. Saroya, 2026 ABCA 20 (dated January 22, 2026), the Court ordered counsel to personally pay $17,500 in costs, opining that:
- The unchecked use of AI resulted in a misleading factum, forcing the respondent to search for non-existent authorities and file additional submissions, and requiring intervention by the Court’s case management officer—all unnecessary and avoidable (paras 8–9);
- The Court’s own time was wasted on addressing the improper use of generative AI in this matter (para 10);
- Delegating drafting to a contractor did not relieve counsel of the obligation to review and correct misleading information, and counsel failed to investigate when concerns were first raised by the respondent (para 12); and
- Counsel improperly and unfair took advantage by adding substantive content to an amended factum when amendments had been directed by the Court’s case management officer to correct the non-existent authorities (para 13).
The Court characterized the conduct as serious misconduct—a marked departure from reasonable professional standards that interfered with the administration of justice (para 16).
The misconduct did not lie in the use of generative AI itself. Neither the LSA Code of Conduct nor the Court’s October 2023 Notice prohibits its use, but both required caution and due diligence. The misconduct arose from counsel’s failure to review and verify a factum for which he bore ultimate responsibility, in turn further compounded by his failure to conduct even a cursory review after concerns were raised. The issue was not the usage of AI, but a failure to meet the professional duty of diligence and the obligation not to mislead the Court.
Generative AI can be a powerful professional tool. It excels at rapidly digesting and organizing information and presenting it in an accessible way. Used appropriately, it can enhance efficiency and support analytical work. However, it is also dangerously prone to producing incorrect information. In addition to exercising caution to avoid disclosure of confidential or privileged information (for example by using secure and localized generative AI) lawyers and other professionals must rigorously verify all statements generated by AI. In the end, generative AI is a tool: it may assist in drafting and organizing information, but accuracy and accountability remain firmly human responsibilities.
Reddy v. Saroya, 2025 ABCA 322 and Reddy v. Saroya, 2026 ABCA 20
Victor Chan