In a notable decision 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-Montréal, 2026 QCCS 1360, rendered on April 22, 2026, the Superior Court established guidelines for the use of artificial intelligence (AI) in the drafting of arbitral awards and, by extension, judicial decisions.
The Superior Court, in reasons authored by the Honourable Martin F. Sheehan, J.C.S., found that the arbitrator had relied on non-existent doctrinal and jurisprudential references central to his reasoning—pointing to the unsupervised use of an AI tool and an improper delegation of decision-making authority.
The decision establishes a clear standard: AI may assist, but must never replace, the decision-maker. When the unsupervised use of AI compromises procedural integrity or public trust, an arbitral award may be set aside.
The Superior Court’s decision
The case arose from a contractual disagreement between an intermediate residential care resource (Osman), its representative association (ARIHQ), and a healthcare facility (now Santé Québec). The dispute resolution process in place provided for arbitration. One of the parties then filed a motion to set aside the arbitral award.
The Court found that the arbitrator had cited sources of legal doctrine and case law that did not exist. The AI had “hallucinated” several key references in the arbitral award. This was particularly evident in the cited decisions of the Court of Appeal of Québec, which do not actually exist.
The Court analyzed these violations in light of articles 646 and 648 of the Code of Civil Procedure, which set out the limited grounds for setting aside an arbitral award in Québec. Specifically, article 646(3) addresses non-compliance with the procedure for appointing an arbitrator or with the applicable arbitration procedure.
The Court clarified that, while AI is not explicitly referenced by article 646 of the Code of Civil Procedure, a substantial breach of the agreed procedure that compromises the integrity of the process may justify the annulment of an arbitral award. In this case, the Court found that the non-existent decisions had a significant impact on the arbitrator’s reasoning, and that he had effectively delegated his decision-making authority to an AI tool. Based on this, the Court set aside the arbitral award, ruling that the non-existent decisions were [translation] “central to the arbitrator’s reasoning.”
However, in setting aside the award, the Court was careful not to establish a general prohibition on the use of artificial intelligence.
Guiding principle for decision-makers regarding the use of AI: Assistance is permitted, but delegation is prohibited
In its decision, the Court established guardrails for the use of AI by arbitrators and other decision-makers. With respect to arbitration specifically, the Court commented on (i) the importance of party autonomy in selecting an arbitrator, (ii) the importance of written reasoning to ensure an informed decision and (iii) the imperative of maintaining public confidence in the arbitration process. The Court concluded that arbitral awards must be authored by the arbitrator chosen by the parties and must not be delegated to third parties.
While the decision does not prohibit or discourage legal professionals or decision-makers from using AI, it establishes guidelines for its use by reiterating certain fundamental principles:
- Delegation of decision-making authority: The power to decide belongs to the decision-maker chosen by the parties, and as such it cannot be delegated to third parties, including AI. The Court identified personal reasoning as a key element of procedural integrity. Using AI without sufficient verification is incompatible with the requirement for the decision-maker to be the author of the reasons underlying the decision.
- The significance of the breach: The Court clarified that using AI as a drafting tool, or even citing erroneous references, will not automatically result in an arbitral award being annulled. In this case, the annulment of the award was due to the central role the non-existent decisions played in the arbitrator’s reasoning, and the fact that the breach was likely to [translation] “affect the parties’ confidence in the outcome and in the arbitration system in general.”
Note: By analogy, the Court ruled on the use of tribunal secretaries by arbitrators, and of law clerks by decision-makers. The same principle was applied: that the participation of such third parties in the decision-making process must not [translation] “undermine the integrity of the process, and the decision-maker must retain responsibility for the drafting.”
Finally, the Court emphasized that other challenges and risks associated with the use of AI must be considered in a judicial or arbitration context. These include:
- the creation of false references or hallucinations;
- the inability to exercise discretion and consider human values;
- biases reflected in AI systems; and
- the lack of privacy associated with certain tools.
Conclusion
This is one of the first decisions worldwide to set out guidelines for how a decision-maker, such as an arbitrator, can use AI. It has already attracted significant international attention.
It represents a broadening of the grounds for annulment based on procedural compliance under article 646(3) of the Code of Civil Procedure, by extending them to cover the use of AI. The Court’s guardrails strengthen arbitration law in Québec by ensuring the quality of arbitral awards and contributing to the development of best practices to guide decision-makers’ use of AI.
