- within Litigation and Mediation & Arbitration topic(s)
- with Senior Company Executives, HR and Inhouse Counsel
- in China
- with readers working within the Insurance and Healthcare industries
Abstract
On April 22, 2026, the Quebec Superior Court of Canada set aside the arbitral award rendered by arbitrator Michel A. Jeanniot in ARIHQ c. Santé Québec. The Court found that the academic literature and case authorities relied upon in the core legal reasoning of the award either did not exist or were unrelated to the propositions for which they were cited. These authorities therefore constituted AI “hallucinated” references and, more importantly, were the only doctrinal and jurisprudential authorities used to support the arbitrator’s legal conclusions. On that basis, the Court held that the arbitrator had improperly delegated part of his adjudicative authority and had abdicated his duty to independently review the outcome of the award, thereby giving rise to a ground for setting aside under Article 646 of the Quebec Code of Civil Procedure, namely that “the applicable arbitration procedure was not observed”. Based on a review of the factual background of the case and the Court’s reasoning, this article further considers recent developments in international arbitration practice and provides practical recommendations for parties to guard against AI “hallucinations” in arbitral proceedings.
Main Text
In recent years, generative artificial intelligence (“Generative AI”) has entered the legal profession at an unprecedented pace. From legal research, contract review and case summarization to the drafting of pleadings, evidence analysis and even assistance in the drafting of awards, large language models (“LLMs”) are becoming increasingly embedded in the daily work of lawyers, arbitrators and judges. At the same time, the problem of AI “hallucinations” has quickly come to light: such models may generate cases, statutory provisions and academic literature that appear highly realistic in format, wording and even citation structure, but in fact do not exist.1 In existing cases, the issue of AI “hallucinations” has mostly remained at the level of lawyers submitting erroneous materials.2 However, where AI-generated false legal authorities are used to support the legal reasoning of an arbitral award, the consequences are no longer limited to professional risk. They may further implicate deeper issues such as the integrity of the arbitral process, adjudicative independence and public confidence in the administration of justice.
On April 22, 2026, in Association des ressources intermédiaires d’hébergement du Québec (ARIHQ) c. Santé Québec – Centre intégré universitaire de santé et de services sociaux du Centre-Sud-de-l'Île-de-Montréal3 (“ARIHQ c. Santé Québec”), the Quebec Superior Court of Canada set aside, for the first time, an arbitral award rendered by sole arbitrator Michel A. Jeanniot on August 8, 2025, which was found to involve AI “hallucinated” citations.4 The Court held that the academic literature and case authorities relied upon by the arbitrator in the core legal reasoning of the award either did not exist at all or bore no connection to the propositions for which they were cited. Those citations were precisely the only doctrinal and jurisprudential authorities used to support the arbitrator’s legal conclusions. On that basis, the Court found that the arbitrator had improperly delegated part of his adjudicative authority and had abdicated his duty of independent verification. Since the false citations caused substantial harm to the integrity of the arbitral process and were capable of reasonably affecting the outcome of the award, the Court ultimately set aside the arbitral award.5
This case is not only regarded as the first case worldwide in which an arbitral award was set aside due to AI “hallucinated” citations, but also marks the point at which the boundaries of AI use in arbitral proceedings have moved beyond issues of compliance and ethics and formally entered the realm of procedural law and setting-aside review.
I. Case Overview
This case arose from an arbitral dispute concerning remuneration for residential care services. Under the dispute resolution clause agreed by the parties, the dispute was to be resolved through a non-institutional arbitration procedure and was to be heard by a sole arbitrator jointly selected by the parties from the list annexed to the agreement.6 In August 2025, arbitrator Michel A. Jeanniot dismissed all arbitral claims on the ground that the applicants had submitted the notice of disagreement out of time. ARIHQ and Osman7 subsequently applied to the Quebec Superior Court to set aside the award, arguing that the award contained numerous fictitious cases, erroneous citations and non-existent legal literature generated by AI. On April 22, 2026, the Quebec Superior Court held that the false citations constituted the core legal basis relied upon in the arbitral award, and found on that basis that the arbitrator had failed to discharge his duty of independent adjudication. This amounted to a ground for setting aside under Article 646 of the Quebec Code of Civil Procedure, namely that “the applicable arbitration procedure was not observed”. The Court ultimately set aside the arbitral award.
II. Key Issues
Whether the arbitrator’s improper use of Generative AI in the course of rendering the award constituted a significant violation of the arbitral procedure sufficient to justify setting aside the award.
III. The Court’s Analysis
The Court ultimately characterized the case as falling within the ground for setting aside under Article 646(3) of the Quebec Code of Civil Procedure, namely that “the applicable arbitration procedure was not observed”,8 and set aside the award on that basis. Justice Martin F. Sheehan stated that, “even if the legislature did not specify each of the procedural violations capable of justifying the setting aside of an arbitral award, it must be understood that any ‘significant violation of the procedure agreed upon by the parties’ which calls into question ‘the integrity of the arbitral process’ may lead to such setting aside”.9 Accordingly, the key issue in this case was not whether AI had been used, but whether such use had resulted in the arbitrator improperly delegating his adjudicative authority, thereby amounting to a significant violation of the arbitral procedure.
1. The Arbitral Procedure Requires the Award to Be Rendered by the Arbitrator Chosen by the Parties
The Court analyzed the issue by reference to the parties’ choice of the arbitrator, the importance of drafting reasons, the confidentiality of deliberations, and the principle prohibiting further delegation. It emphasized that “decisions should be drafted by the arbitrator chosen by the parties, without delegation to third parties”.10
First, the Court noted that the parties’ agreement to submit their dispute to a particular arbitrator is, in essence, based on their trust in that arbitrator’s personal expertise, experience and independent judgment. As Justice Martin F. Sheehan observed, “allowing the parties to choose their decision-maker is, moreover, a ‘considerable advantage over the ordinary courts’”.11 Accordingly, the parties are entitled to expect that the chosen arbitrator himself will independently assess the facts, evidence and legal issues in the case, rather than having any third party, including Generative AI, substantially substitute for him in forming the reasoning and conclusions of the award.
Second, the Court emphasized that an arbitrator is under a duty to give reasons, and that this duty is not merely formal in nature, but directly bears on the validity of the award, procedural fairness and the parties’ confidence in the process. Justice Martin F. Sheehan recalled that the Quebec Court of Appeal had already confirmed the nullity of an unreasoned arbitral award.12 He further referred to the Supreme Court of Canada’s reasoning in Baker,13 observing that written reasons assist decision-makers in clearly defining the issues and articulating their reasoning, thereby promoting more rigorous and careful legal analysis and assuring the parties that their submissions have been properly considered.14 In other words, the very process of drafting reasons is itself a process through which the arbitrator tests and refines his own judgment. For that reason, the arbitrator’s adjudicative authority should not, in principle, be delegated to another person.
Finally, the Court emphasized the principles of confidentiality of deliberations and the prohibition against further delegation. Article 644 of the Quebec Code of Civil Procedure expressly requires arbitrators to maintain the confidentiality of the deliberative process.15 Referring to the Quebec Court of Appeal’s reasoning, Justice Martin F. Sheehan noted that this obligation “requires them to abstain from communicating with third parties about the matter to be decided once it has been taken under advisement. The ultimate purpose is to ensure that the arbitral award reflects the opinion of the arbitrators alone, and not those of anyone else”.16 In addition, the Court further referred to the principle of delegatus non potest delegare under Canadian law, namely that a person or body vested with adjudicative authority must, in principle, exercise that authority personally and may not further delegate it.17 The Court did not deny the possible use of technological assistance, but made clear that “the participation of third parties in the drafting process must not undermine the integrity of the process, and responsibility for the drafting must remain with the decision-maker”.18 Otherwise, such conduct may amount to an improper further delegation of arbitral decision-making authority.
2. The False Citations in This Case Formed the Core Basis of the Award’s Reasoning and Constituted a Failure to Observe the Arbitral Procedure
In this case, Justice Martin F. Sheehan identified, one by one, the false citations in the arbitral award that had been verified by the Court, and confirmed that the cited materials either did not exist at all or that the corresponding reference numbers in fact led to entirely unrelated cases. For example, the academic article cited in footnote 4, namely Frédéric Bachand, “Prescription et déchéance: frontières mouvantes et enjeux pratiques”, Développements récents en droit des contrats, Barreau du Québec, 2016, could not be found. The cases cited in footnotes 5 to 7 — Ville de Montréal c. Syndicat des cols bleus regroupés de Montréal, 2005 QCCA 591; Groleau et Groupe Pages Jaunes Cie, 2011 QCCS 5386; and Tremblay c. Commission scolaire de la Jonquière, 2002 CanLII 24357 (QCCA) — were found by the Court not to exist, and the numerical references instead led to other unrelated decisions. As for the arbitral precedent cited in footnote 8, Arbitrage CHU Ste-Justine (D.T.E. 2018-30), SOQUIJ, Quebec’s legal information database, confirmed that no such arbitral award existed. The Court further noted that the decision 2018EXPT-30 — Sonin c. Université Concordia, rendered by the Administrative Labour Tribunal — concerned a complaint dismissed because it had been filed outside the 30-day statutory time limit provided under the Labour Code. It therefore did not concern a contractual limitation or forfeiture period, and could not support the arbitrator’s proposition that “an arbitral tribunal recognized a 30-day contractual time limit”.19
The Court expressly noted that these citations were not peripheral oversights in the award. Rather, “the references mentioned above are at the heart of the arbitrator’s reasoning. They constitute the only doctrinal or case-law references used as legal support for the Award.”20
Precisely because these false citations lay at the heart of the award’s reasoning, the Court found that the evidence was sufficient to conclude that the arbitrator had improperly delegated his decision-making authority and had abdicated his duty to independently review the outcome of the award. This was likely to have had a significant impact on the result of the award and therefore constituted a serious procedural violation requiring the award to be set aside.
However, Justice Martin F. Sheehan also made clear that the case does not mean that any use of AI to assist in the drafting of an award, or any erroneous citation in an award, will automatically result in setting aside. Whether setting aside is justified must still be assessed on a case-by-case basis, taking into account “the nature of the breach in light of the arbitral procedure engaged, the determination of the impairment to the integrity of the procedure, and the assessment of the impact of the breach on the award”.21 The reason why the threshold for setting aside was met in this case was that the false citations were not isolated or peripheral errors, but lay at the core of the arbitrator’s legal reasoning.
IV. Practical Recommendations
1. Verification of and Responses to Awards Containing AI “Hallucinated” Citations
The Court also noted in its decision that AI technology should not, in itself, be categorically excluded. However, its involvement in arbitral and judicial proceedings nevertheless gives rise to challenges and risks that cannot be ignored.
On that basis, the Court identified five risks associated with the use of AI in legal decision-making contexts. The first is the risk of “hallucinations”, namely that AI may generate legal citations that appear realistic in form but are false in substance, and that such citations may be difficult for both parties and decision-makers to identify in a timely manner. The second is the limitation of AI in exercising discretion and judgment: judicial or arbitral decision-making often requires consideration of community values and the circumstances of the particular case, whereas AI operates according to predetermined logic and is generally unable to properly address value judgments that vary depending on the persons and circumstances involved. The third is the risk of algorithmic bias, as AI may produce systemic bias due to defects in training data or program design, while its operation may be difficult to supervise effectively because of its opacity. The fourth is the risk to confidentiality: case information entered into an AI platform may be incorporated into the platform’s database and later disclosed to other users, which may in turn violate the duty to maintain the confidentiality of deliberations. The fifth is the risk to public confidence: if the public comes to believe that decisions are made by opaque machines rather than by human decision-makers, the public confidence on which the judicial and arbitral systems depend may be fundamentally undermined.22
Therefore, upon receiving an arbitral award, parties should carefully review the reasons for the award, with particular attention to the statutory provisions, cases and academic literature relied upon by the tribunal to support its core legal conclusions. For key citations, parties should not assess their authenticity merely on the basis of the case name, citation number or summary set out in the award. Instead, they should cross-check them against the original judgments, official databases and authoritative legal databases to verify whether the cited materials actually exist, whether the citation numbers correspond to the correct cases, and whether the content cited in the award is consistent with the original sources.
Once a party discovers fictitious cases, erroneous citations or non-existent academic literature generated by AI “hallucinations” in an award, it should immediately preserve the relevant evidence, including the original award, database search results, search paths, webpage screenshots and materials showing that the relevant citations could not be located, and should promptly pursue appropriate procedural remedies.
Where such false citations lie at the core of the award’s legal reasoning and may have affected the outcome of the award, they may serve as important evidence in an application to set aside the award, on the basis that there was a significant defect in the arbitral procedure that impaired the integrity of the process. In jurisdictions that have adopted the UNCITRAL Model Law on International Commercial Arbitration, a party may rely on Article 34(2)(a)(iv) to apply for the setting aside of an award on the ground that “the arbitral procedure was not in accordance with the agreement of the parties”.23 Under Chinese law, a party may apply to set aside an award pursuant to Article 71(3) of the Arbitration Law of the People’s Republic of China, which provides that an award may be set aside where “the arbitration procedure violates statutory procedures”.24
Where the opposing party further seeks recognition and enforcement of an arbitral award containing false citations, the affected party may also raise objections under the applicable law. In cases where the New York Convention applies, a party may rely on Article V1(d) and request the court to refuse recognition and enforcement of the award on the ground that “the composition of the arbitral authority or the arbitral procedure was not in accordance with the agreement of the parties, or, failing such agreement, was not in accordance with the law of the country where the arbitration took place”.25
2. Clarifying Rules on the Use of Artificial Intelligence in Arbitral Proceedings
In 2025, the Chartered Institute of Arbitrators (“CIArb”) issued its Guideline on the Use of AI in Arbitration, together with a template agreement and a template procedural order on the use of AI in arbitration,26 providing relatively detailed practical guidance for prior agreement, procedural management and risk control regarding the use of AI in arbitral proceedings. The China International Economic and Trade Arbitration Commission (“CIETAC”) has also recommended, in its Guidelines on the Use of Artificial Intelligence Technology in Arbitration, that parties may agree on the use of AI in their arbitration agreement, and that the arbitral tribunal may seek the parties’ views on the use of AI in a procedural order or at a pre-hearing conference.27
From a practical perspective, parties may consider agreeing in advance on rules governing the use of AI in the arbitration agreement, or having such rules further confirmed or refined by the arbitral tribunal in a procedural order or at a pre-hearing conference. Such rules may include, without limitation: prohibiting the arbitral tribunal from delegating substantive legal analysis or adjudicative authority to AI; prohibiting parties or their counsel from submitting AI-generated cases, citations or factual content without verification; prohibiting the uploading of confidential materials, trade secrets or non-public evidence to AI platforms; and establishing a disclosure mechanism for the use of AI, where necessary, requiring disclosure of the name of the AI tool, the purpose of its use and the scope of its use.
Footnotes
1 Matthew Dahl, Varun Magesh, Mirac Suzgun, Daniel E Ho, Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models, Journal of Legal Analysis, Volume 16, Issue 1, 2024.
2 See Mata v. Avianca, Inc., No. 22-cv-1461 (PKC) (S.D.N.Y. 2023); Ren v. Cao and Beijing International Travel Agency Co., Ltd., a dispute related to a company, Beijing Tongzhou District People’s Court, (2024) Jing 0112 Min Chu No. 19067, People’s Court Case Database Entry No. 2025-18-2-494-001.
3 Association des ressources intermédiaires d'hébergement du Québec (ARIHQ) c. Santé Québec - Centre intégré universitaire de santé et de services sociaux du Centre-Sud-de-l'Île-de-Montréal, 2026 QCCS 1360 (Can. Que. Sup. Ct. Apr. 22, 2026). (https://endroit.ca/wp-content/uploads/2026/06/Jugement-2026-Qccs-1360-Pdf.pdf)
4 Supra note 3, para. 122.
5 Supra note 3, para. 120-121.
6 Supra note 3, para. 16、26.
7 ARIHQ, short for Association des ressources intermédiaires d’hébergement du Québec, is a recognized representative organization for intermediate residential resources in Quebec. Osman is one of its member institutions and primarily provides residential care services for adults. Under the National Agreement applicable to residential care services, ARIHQ is entitled to represent its member institutions in dispute resolution proceedings. Accordingly, ARIHQ and Osman jointly acted as applicants in the arbitration and in the subsequent setting-aside proceedings.
8 Code of Civil Procedure
646. The court cannot refuse to homologate an arbitration award or a provisional or safeguard measure unless it is proved that
3° the procedure for the appointment of an arbitrator or the applicable arbitration procedure was not observed.
9 Supra note 3, para. 71.
10 Supra note 3, para. 86.
11 Supra note 3, para. 73.
12 Supra note 3, para. 77.
13 Baker c. Canada (Ministre de la Citoyenneté et de l'Immigration), [1999] 2 RCS 817.
14 Supra note 3, para. 79.
15 Code of Civil Procedure
644. The arbitrator is required to preserve the confidentiality of the arbitration process and protect deliberative secrecy but violates neither by stating conclusions and reasons in the award.
16 Supra note 3, para. 83.
17 Supra note 3, para. 85.
18 Supra note 3, para. 87.
19 Supra note 3, para. 103-112.
20 Supra note 3, para. 113.
21 Supra note 3, para. 117-118.
22 Supra note 3, para. 90.
23 UNCITRAL Model Law on International Commercial Arbitration 1985 (With amendments as adopted in 2006)
Article 34. Application for setting aside as exclusive recourse against arbitral award
(2) An arbitral award may be set aside by the court specified in article 6 only if:
(a) the party making the application furnishes proof that:
(iv) the composition of the arbitral tribunal or the arbitral procedure was not in accordance with the agreement of the parties, unless such agreement was in conflict with a provision of this Law from which the parties cannot derogate, or, failing such agreement, was not in accordance with this Law.
24 Arbitration Law of the People’s Republic of China (2025)
Article 71 Where a party submits evidence to prove that the award has any of the following circumstances, it may apply to the intermediate people’s court at the place where the arbitration institution is located to set aside the award:
(3) The composition of the arbitral tribunal or the arbitration procedure violates statutory procedures.
25 Convention on the Recognition and Enforcement of Foreign Arbitral Awards (the “New York Convention”, 1958)
Article V
1. Recognition and enforcement of the award may be refused, at the request of the party against whom it is invoked, only if that party furnishes to the competent authority where the recognition and enforcement is sought, proof that:
(d) The composition of the arbitral authority or the arbitral procedure was not in accordance with the agreement of the parties, or, failing such agreement, was not in accordance with the law of the country where the arbitration took place.
26 CIARB Guideline on the Use of AI in Arbitration (2025).
27 CIETAC Guidelines on the Use of Artificial Intelligence Technology in Arbitration (Trial)
V. Recommended Checklist for Reducing Risks in the Use of Artificial Intelligence Technology
The following measures may reduce the risks associated with the use of artificial intelligence technology:
5.1 Unless otherwise provided by law, parties may incorporate provisions concerning the use of artificial intelligence into their arbitration clauses.
5.2 Where necessary, the arbitral tribunal may invite the parties, in a procedural order or at a pre-hearing conference, to express their views on the use of artificial intelligence, including but not limited to which artificial intelligence tools may be used, restricted or prohibited, and the reasonable disclosure of circumstances involving the use of artificial intelligence.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]
