In a decision published on the Kluwer Arbitration Blog on 19 June 2026, the Québec Superior Court became the first state court to set aside an arbitral award on the ground that it was drafted with the help of artificial intelligence and rested on authorities that the technology had invented. The case, ARIHQ v Santé Québec, confronts directly a question that arbitration practice has so far only debated in the abstract: when does an arbitrator's reliance on AI-generated material so undermine the integrity of an award that it must be annulled.
The dispute arose under a 2018 National Agreement between the Association des ressources intermédiaires d'hébergement du Québec (ARIHQ), a recognised representative organisation, and the Minister of Health and Social Services, concluded on behalf of an accommodation service provider, Centre de Santé Osman. The agreement set out a multi-tier dispute resolution procedure under which any disagreement had to be raised by written notice within a peremptory period of 90 days, failing amicable settlement, before being submitted to arbitration. A 2021 dispute over retroactive payments led ARIHQ, almost three years later, to claim roughly CAD 1,225,000 from the respondent, Santé Québec, a public health and social services authority. Santé Québec rejected the claim as time-barred, and the arbitrator dismissed the case, holding the 90-day deadline valid and not contrary to public policy.
ARIHQ and Osman then applied to annul the award before the Superior Court on two grounds: that it gave effect to a contractual limitation period shorter than the statutory three-year period, and that the arbitral procedure had not been respected because there was strong evidence the award had been drafted by AI. It was the second ground that the court addressed in the terms now drawing attention across the arbitration community.
The application proceeded under paragraph 3 of Article 646 of the Québec Code of Civil Procedure, which permits annulment where the applicable arbitral procedure was not observed. The court held that a procedural violation must be significant and must undermine the integrity of the arbitral process before judicial intervention is warranted; an insignificant irregularity causing no prejudice or effect on the outcome does not suffice. It then anchored its analysis in party autonomy: those who choose arbitration are entitled to expect that the arbitrator they selected will personally render a reasoned decision and will not delegate the core decision-making function to a third party.
The court acknowledged AI's potential benefits but identified its principal risks, including hallucinated material, the inability to exercise genuine discretion, hidden and undetectable bias, threats to confidentiality, and erosion of public confidence in justice. On that basis it held that a decision-maker using AI bears stricter duties than counsel, because the legitimacy of a decision depends on the reasoning supplied by the decision-maker rather than on external rationalisation. Examining the award, the court found that the arbitrator had relied on hallucinated case law and non-existent doctrinal sources that were central to his reasoning. By failing to verify those authorities, he had in effect delegated his authority. The court set the award aside and ordered the parties to appoint a new arbitrator within 60 days.
The reasoning carries weight beyond Québec because the relevant provision closely mirrors Article V(1)(d) of the New York Convention and Article 34(2)(a)(iv) of the UNCITRAL Model Law, the Code having been reformed in 1986 along those lines. Courts applying those instruments generally require a substantial procedural defect or a causal link between the defect and the award. Here that link was direct: because the fabricated authorities formed a central part of the reasoning, the applicants could reasonably believe the arbitrator might have decided differently had the citations been verified.
The decision does not impose a blanket prohibition on AI use by arbitrators. The court was explicit that not every award containing erroneous citations or AI assistance must be annulled; independent verification, or a merely minor irregularity, would not have justified setting the award aside. What it confirms is narrower and firmer: AI may assist but cannot replace the arbitrator, and the responsibility to render a reasoned award reached by an independent human decision-maker remains with the person the parties chose.