You used AI instead of a lawyer? Good Luck!

The Use of AI Before the Courts

Artificial intelligence has gone from novelty to necessity in remarkably little time. What felt futuristic only a few years ago has become a daily companion for lawyers, judges, and self-represented litigants alike. For many, drafting a pleading, researching case law, or even preparing reasons for judgment without the assistance of AI has become almost unimaginable.

There is nothing inherently troubling about that. Used responsibly, generative AI has the potential to improve efficiency, reduce costs, and enhance access to justice.

AI and its limitations and dangers

The problem arises when convenience replaces judgment. Increasingly, courts are confronted with pleadings, briefs, and other judicial documents generated largely by AI and filed with little, or no, verification of their contents. The results are becoming all too familiar: fabricated case law, non-existent statutory provisions, invented quotations, and legal principles that exist only in the imagination of a language model.

How courts are responding to AI hallucinations in legal proceedings.

Quebec courts are taking notice. While judges have consistently recognized that AI can be a valuable tool and have been careful not to discourage its responsible use, they have also made one point unmistakably clear: artificial intelligence is no substitute for legal judgment. Whether a document is drafted by a lawyer, a self-represented litigant, or ChatGPT itself, the person who signs and files it remains fully responsible for its accuracy. The growing body of case law leaves little doubt that those who fail to verify AI-generated content risk more than embarrassment - they risk sanctions.

Courts are increasingly responding to this phenomenon by imposing significant sanctions on parties who misuse artificial intelligence or file AI-generated material without proper verification. Recent decisions of the Superior Court of Québec illustrate this growing trend:

• Specter Aviation Limited c. Laprade, 2025 QCCS 3521: The defendant admitted relying on AI to prepare his defence. The pleading contained numerous fictitious authorities. Justice Morin emphasized that AI should not be stigmatized, but its use does not relieve litigants of their obligation to verify the accuracy of court filings. Because the filing wasted judicial resources, the defendant was ordered to pay $5,000 under article 342 C.C.P.

 

• Entreprises Bertrand Roberge ltée c. Giroux, 2025 QCCS 4157 : This case demonstrates that even judges are not immune from the risks associated with AI. Justice Geoffroy cited a fictitious Supreme Court authority and several unrelated decisions in support of the doctrine of piercing the corporate veil. The judgment also contained purported verbatim quotations that appeared nowhere in the evidentiary record. While isolated citation errors can occur, the number and nature of the inaccuracies suggested the improper use of generative AI.

 

• Bourse de l’Immobilier Multilogements inc. c. Lanthier, 2025 QCCS 4135 and Kalala c. Coopérative d'habitation La Fraternité Micheloise, 2025 QCCS 4866 : Both decisions involved pleadings prepared with the assistance of AI that contained fictitious legal authorities. In Kalala, the plaintiff also relied on provisions of the Civil Code of Québec dealing with emancipation that bore no relation to the issues before the Court, further suggesting AI-generated inaccuracies. In each case, the Court held that filing fabricated authorities—even absent an intention to mislead—constitutes a serious procedural breach because it undermines the integrity of the judicial process and forces both the Court and the opposing party to undertake unnecessary verification. As a result, the plaintiff in Kalala was ordered to pay $500 in compensation under article 342 C.C.P., while the party in Bourse de l'Immobilier was ordered to pay $750.

 

• Alamleh c. R., 2026 QCCS 416: The Court recognized AI as a valuable tool but emphasized that users remain personally responsible for verifying its output. The applicant's submissions contained fictitious authorities, a non-existent statute and incorrect citations, all of which weighed against his application.

 

• Droit de la famille — 251792, 2025 QCCS 4505 : The Court addressed the litigant's reliance on fictitious authorities and inaccurate representations of the Code of Civil Procedure. The litigant acknowledged having prepared with ChatGPT, and the Court noted that several cited decisions did not exist and that the statutory provisions relied upon either failed to support, or directly contradicted, his arguments. Although the Court concluded that a warning was sufficient in the circumstances, it emphasized that the false citations had unnecessarily consumed both the opposing party's and the Court's time, cautioning that repeated conduct of this nature could give rise to a finding of abuse of process.

 

• Droit de la famille — 26184, 2026 QCCS 666 : The Court found that the applicant's filings contained fabricated legal authorities, incorrect statutory references and invented quotations falsely attributed to counsel and child protection officials. After acknowledging the use of AI, the applicant was warned that such conduct had misled both the Court and opposing counsel and could justify a future finding of abuse of process.

 

• Blinds to Go Inc. c. Blachley, 2025 QCCS 3190: Cases referenced by one of the parties do not seem to exist. Justice Lukasz Granosik noted that the cited authorities appeared to be AI "hallucinations" with no basis in law.

 

•  Felipe Coimbra c. Doria Boukheroufa et. al., File no 500-11-064871-243: In two unpublished Superior Court decisions arising from the same court file^[1] in which our firm represents the defendants, two of the plaintiff’s successive AI-generated proceedings were declared abusive and were dismissed for reason of lacking any legislative or jurisprudential foundation, resulting in his condemnation to pay damages totaling $2,500.

However, the wrongful and/or abusive use of AI in court proceedings has also been observed in other cases before the Court of Appeal^[2], the Court of Quebec^[3], the Tribunal administratif du logement^[4], the Tribunal administratif du travail^[5], the Tribunal administratif^[6], the Régie du bâtiment^[7], and labour arbitration awards^[8].

While the factual circumstances differ, the message from the courts is remarkably consistent. These decisions reveal a clear trend. Quebec courts, as well as courts across Canada^[9] and worldwide^[10], are not condemning the use of AI itself. Rather, they are sanctioning the uncritical submission of AI-generated material. The obligation to verify legal authorities remains a personal responsibility that cannot be delegated to technology.

Although these issues affect all participants in the justice system, self-represented litigants appear to be particularly vulnerable. Indeed, more than 80% of documented instances of abusive AI use in legal proceedings have involved self-represented parties^[11]. This is perhaps unsurprising. Without legal training or the benefit of professional oversight, self-represented litigants may be less equipped to identify fabricated authorities, inaccurate legal propositions, or fictitious quotations generated by AI. While this reality underscores the potential of AI to improve access to justice, it also illustrates the risks of relying on such tools without careful verification and a sound understanding of the applicable law.

A pleading is not merely another document generated on a computer. It is a formal representation made to the Court, signed by a litigant or counsel, and submitted in support of rights, obligations, and legal consequences. The filing of judicial proceedings is therefore a solemn act that demands diligence, accuracy, and intellectual honesty. Those obligations cannot be delegated to a language model.

Artificial intelligence will undoubtedly continue to transform legal practice, and it should. Used responsibly, it is a valuable tool that can improve efficiency and expand access to justice. But AI cannot exercise professional legal judgment, develop litigation strategy, or assume responsibility for the documents filed with a court. What may appear to save time at the outset often results in greater costs once fabricated authorities, inaccurate legal propositions, and procedural defects must be identified and corrected.

Conclusion

As recent case law demonstrates, what appears to save time today can quickly become tomorrow's procedural headache. It remains advisable, despite the advancement of technology, to consult an attorney to seek advice and/or representation.

Author: Audrey Baillairgé  

[1] The first is in annex to the minutes of the hearing of June 4, 2025 (unpublished judgement rendered by Martin F. Sheehan) and the second is a transcript of the judgment delivered orally and digitally recorded in the Court on May 6, 2025 (unpublished judgement by David R. Collier).

[2] Azar c. Université de Montréal, 2025 QCCA 1442 at para 7.

[3] Lessard c. Longuépée, 2025 QCCQ 8285 at para 38; Makongo v. Montpetit, 2026 QCCQ 746 at para 41.

 [4] Rajabi v. Lassalle, 2024 QCTAL 35477 at para 18 [*]; Thabize c. Immobilière montérégienne IMR inc., 2026 QCTAL 7284 at para 20.

 [5] St-Roch c. Andritz Hydro Canada inc., 2025 QCTAT 5136 at para 46 [*]; Bégin-Létourneau c. Syndicat des spécialistes et professionnels d’Hydro-Québec, section locale 4250, SCFP-FTQ, 2025 QCTAT 5208 at para 20.

[6] Dulac c. Ville de Gatineau, 2025 CanLII 113234 (QC TAQ) at para 52 [*].

[7] Régie du bâtiment du Québec c. 9308-2469 Québec inc. (Éco résidentiel), 2025 QCRBQ 86 at para 162.

[8] Syndicat des travailleuses et travailleurs de L’Autre Maison – CSN c Centre L’Autre Maison inc., 2025 CanLII 120680 (QC SAT) at para 125.

[9] Zhang v. Chen, 2024 BCSC 285 : citation de jurisprudence inexistante par un avocat ayant reconnu s’être fié sur les bons conseils de ChatGPT était constitutive d’un abus de procédure et était de nature à remettre l’intégrité du système de justice en question (para 29 et 46).  Voir aussi: Hussein v. Canada (Immigration, Refugees and Citizenship), 2025 FC 1138, par. 9. Lloyd's Register Canada Ltd. v. Choi, 2025 FC 1233 : rejet de la procédure entachée de références « hallucinées » et générées par l’intelligence artificielle.

[10] In November 2023, a Brazilian judge issued a decision containing numerous errors and fictitious legal authorities after relying on AI-generated content. When called upon to explain, the judge stated that drafting had been delegated to a staff member, who in turn had used ChatGPT. In October 2025, two U.S. federal judges were formally reprimanded after filing judicial documents containing fabricated citations and fictitious passages. Each attributed the errors to a member of their legal support staff rather than acknowledging personal use of AI.

[11] The Rise of AI-Hallucinated Case Law in Canadian Courts and Tribunals, 2026 CanLIIDocs 738, online: <https://www.canlii.org/en/commentary/doc/2026CanLIIDocs738?resultId=undefined&searchId=2026-07-27T13:27:55:336/015b25540dfb47e392ae3f33024bdc54#!fragment//BQCwhgziBcwMYgK4DsDWszIQewE4BUBTADwBdoByCgSgBpltTCIBFRQ3AT0otokLC4EbDtyp8BQkAGU8pAELcASgFEAMioBqAQQByAYRW1SYAEbRS2ONWpA> .