In Association des ressources intermédiaires d’hébergement du Québec (ARIHQ) v. Santé Québec – Centre intégré universitaire de santé et de services sociaux du Centre-Sud-de-l’Île-de-Montréal, 2026 QCCS 1360, the Court annulled an arbitral award after concluding that the arbitrator had relied on non-existent doctrine and case law seemingly generated through AI. The Court found that the reliance on hallucinated authorities undermined the integrity of the arbitral process and justifed annulment under articles 646 and 648 of the Code of Civil Procedure of Quebec (C.C.P.) The issue was not whether the use of artificial intelligence was itself a ground for setting aside the arbitral award; the question was whether the arbitral procedure chosen by the parties had been respected. The Court emphasized that its decision should not be understood as prohibiting the use of AI by arbitrators, but rather as requiring that the arbitrator remain personally responsible for the reasoning and authorities supporting the award.
Background – The members of the Plaintiff ARIHQ (“the Association”) were intermediate resources in the Quebec public health and social services network comprised of privately operated facilities that provided accommodation, support, and care to their adult residents. The Association negotiated with the Ministry of Health and Social Services (“the Ministry”) on behalf of its members and, in 2018, they concluded a National Agreement governing (in particular) the remuneration scale applicable to the Association’s members when providing services. The National Agreement contained an arbitration clause.
Ultimately, a dispute arose after one member of the Association (“Osman”) sought retroactive compensation from the public health institution to which it provided services, the Centre intégré universitaire de santé et de services sociaux du Centre-Sud-de-l’Île-de-Montréal (“the Montréal Centre”). Nearly three years after the Montréal Centre rejected Osman’s claim, the Association delivered a notice of disagreement and subsequently referred the dispute to arbitration.
The Montréal Centre sought a preliminary dismissal of the arbitration on the basis that the notice of disagreement had been filed outside the contractual time limits. The arbitrator agreed and dismissed the claim without hearing the merits, concluding that the contractual deadlines were valid and had not been respected.
The annulment application– The Association and Osman applied to annul the award on two principal grounds. First, they argued that the award was contrary to public order because it effectively enforced a contractual prescription period shorter than the imperative three-year period established by the Civil Code of Québec. Second, they argued that the arbitral procedure had not been respected because the award cited numerous doctrinal writings and judicial authorities that simply did not exist, suggesting that the reasons had been generated, at least in part, through the use of generative AI and that the arbitrator had consequently delegated his decision-making responsibility.
The decision – The Court considered two issues: (i) the breach of public order argument on the substantive issue; and (ii) the allegation that the arbitrator had used AI in drafting the award and the argument that this was a breach of fair process.
(i) Public order – The Court rejected the public order argument under the second paragraph of Article 646 C.C.P. It reiterated that an application to annul is not an appeal and that courts cannot review the correctness of an arbitrator’s interpretation of the law. Even if the arbitrator had erred in distinguishing contractual notice requirements from prescription, such an error would concern the merits of the dispute rather than the limited grounds for annulment.
(ii) Procedural irregularity and AI – The Court accepted the Plaintiffs’ second ground of annulment. It observed that the issue is not whether the use of artificial intelligence was itself a ground for setting aside an arbitral award. Rather, the question was whether the arbitral procedure chosen by the parties had been respected, failing which annulment can be sought pursuant to Article 646 (3) C.C.P.
The Court emphasized the primary importance of the will and autonomy of the parties, which underlies the ground for annulment associated with a failure to respect the procedures chosen by the parties for the appointment of the arbitrator and the conduct of the arbitration. To the extent that these choices are not respected and the breach of procedure is determined to be significant, the validity of the award itself can be compromised.
The Court linked this principle to several fundamental features of arbitration:
- party autonomy in selecting the arbitrator;
- the importance of reasoned awards in demonstrating that the arbitrator independently analyzed the issues submitted by the parties; and
- the arbitrator’s obligation to maintain the secrecy of deliberations and the prohibition against delegating the decision-making function.
Taken together, these factors support the expectation that the arbitrator must personally exercise the adjudicative function entrusted to them.
The Court stressed that these principles do not prohibit every use of AI. It expressly acknowledged that modern AI tools may legitimately assist legal professionals with certain tasks, but cautioned against the inherent challenges and risks that such use presents, specifically:
- False references or hallucinations of legal authorities;
- The absence of human values or judgment and the role of discretion in decision-making;
- Bias in AI-generated content and lack of transparency in the statistical basis in the underlying programming;
- Confidentiality risks; and
- Public confidence in the justice system.
These considerations, generally examined from the perspective of legal professionals in their submissions to the courts, have been raised in relation to the potential negative impact on the resulting decision-making. When the contemplated use is by the decision-makers themselves, the risks are even more significant. In this respect, the Court cited the Canadian Judicial Council’s Guidelines for the Use of Artificial Intelligence in Canadian Courts, which explicitly prohibit the delegation of decision-making authority to computer programs.
The Court then examined the award itself, reviewing a series of doctrinal and case law sources cited by the arbitrator to support his legal conclusions that either did not exist or corresponded to an entirely unrelated decision. Those authorities were not peripheral references. Rather, they constituted the only authorities cited in support of the arbitrator’s legal analysis. The remaining citations in the award merely reproduced authorities advanced by the parties themselves.
On that record, the Court concluded, on a balance of probabilities, that the arbitrator had effectively abdicated part of the decision-making function entrusted to him.
Finally, the Court was careful to limit the scope of its decision. It expressly declined to hold that every use of AI, or every erroneous citation, would justify an annulment. Whether a procedural irregularity warrants intervention remains a contextual inquiry requiring consideration of the seriousness of the defect, its impact on the integrity of the arbitral process, and whether it likely affected the result.
In this case, because the fabricated authorities formed the core of the arbitrator’s legal reasoning, the Court concluded that the procedural breach was substantial and likely influenced the outcome:
“[120] [TRANSLATION] In the present case, the irregularity is significant. It is capable of undermining the parties’ confidence in both the award and the arbitral process more generally. Since the non-existent decisions lie at the heart of the arbitrator’s reasoning, a party could reasonably conclude that a more thorough verification of those decisions might have led the arbitrator to reconsider his position.”
The award was thus annulled and the parties were directed to appoint a new arbitrator.
Commentary:
First, and importantly, this is not a judicial rejection of artificial intelligence in arbitration. To the contrary, the Court expressly acknowledged that AI can legitimately assist arbitrators and legal professionals with research, document review, drafting assistance, translation and other administrative tasks. That said, this judgment and the facts giving rise to it, highlight the growing need for clear guidelines governing the use of AI by arbitrators, much as courts, law societies and professional regulators have begun to develop for lawyers and judges. The Chartered Institute of Arbitrators (Ciarb) Guideline on the Use of AI in Arbitration provides a good example of direction for arbitrators that reflects the concerns raised by the Court regarding the importance of ensuring that the decision is the result of the arbitrator’s judgment:
“8.2 Arbitrators should not relinquish their decision-making powers to AI but may use AI to support more accurate and efficient processing of submitted information, always ensuring independent judgement. Arbitrators are advised to refrain from using AI in ways that could compromise the integrity of the proceedings or the validity or enforcement of the award. Specifically, the Tribunal should avoid delegating any tasks to AI Tools, such as legal analysis, research and interpretation of facts and law, or application of the law to the facts, if such use could influence procedural or substantive decisions”.
This guidance reflects the central concern underlying the Court’s decision, namely the importance of ensuring that the decision is the result of the arbitrator’s judgment. Adhering to guidelines like this can serve to maintain confidence in arbitration without discouraging the responsible adoption of technologies that have the potential to improve both efficiency and access to justice.
Second, party autonomy, the cornerstone of arbitral jurisdiction, is the central element of the Court’s reasoning in this case and is the prevalent concern when considering the use of AI by arbitrators. As the Court emphasized in the decision, the parties choose the specific arbitrator(s) to decide on their dispute. That choice necessarily informs the permissible scope of AI in the arbitral process. While an arbitrator cannot delegate the decision-making power entrusted to them by the parties, the parties themselves may choose to entrust part, or even all, of that power to an AI-based solution. They may do so for reasons of cost, speed, or efficiency. Party autonomy suggests that the parties themselves are free to agree to a different model. For example, the AAA-ICDR (American Arbitration Association-International Centre for Dispute Resolution) has developed an AI arbitrator platform that is available to issue awards in two-party, document-only construction disputes under a specified monetary threshold.
That said, parties contemplating AI-driven dispute resolution should remain mindful of the concerns identified by the Court: bias, a lack of transparency, and the potential inability to understand or scrutinize how generative AI reaches its conclusions. Those concerns may ultimately undermine the very party autonomy that may be invoked to justify the use of AI in the first place. These concerns are echoed in the comments of a recent Speech by the UK’s Master of the Rolls on the subject of AI in arbitration, in which he raised concerns about the impartiality and independence an AI arbitrator, while nevertheless recognizing that party autonomy remains the guiding principle.
Third, this judgment leaves open an intriguing question: what grounds remain available to challenge an award where the parties have expressly agreed that AI will perform some, or all, of the decision-making function?
In this case, the award was annulled because the arbitrator’s apparent reliance on AI undermined the parties’ agreement that he would decide the dispute, thereby resulting in a failure to respect the applicable arbitration procedure. In Québec, that procedural defect engaged article 646(3) C.C.P. Similar grounds exist under other Canadian domestic arbitration statutes, which permit challenges based on procedural fairness. For international arbitration, annulment could be sought for the failure to respect the agreed-upon composition of the arbitral tribunal or applying an arbitral procedure that is not in accordance with the parties’ agreement (UNCITRAL Model Law, article 34(2)(a)(iv)). If, however, an AI arbitrator or the use of AI for decision-making is requested and sanctioned by the parties, this decision suggests that there will be very limited room to seek the annulment of the resulting award.
Finally, this decision, and the questions that remain unanswered, underscore the practical importance of considering AI expressly when drafting arbitration clauses. As AI becomes increasingly integrated into every stage of the arbitral process, parties should consider whether, and to what extent, AI may be used by arbitrators (or the tribunal secretary), whether any disclosure obligations should apply, and whether specific safeguards or limitations are appropriate. While these issues can readily be addressed in new arbitration clauses, parties should also consider whether existing arbitration agreements warrant amendment. Clearly defining the permissible scope of AI use at the outset will better ensure that the arbitral process reflects the parties’ intentions while reducing uncertainty and, potentially, the scope for future challenges to the resulting awards.