The Québec Superior Court has set aside an arbitration award in ARIHQ v. Santé Québec after discovering that the arbitrator relied on legal authorities generated by artificial intelligence — cases that do not exist.

The ruling, issued in the Province of Québec, Canada, is a landmark decision: it is believed to be the first time a court in the province — and potentially in Canada — has vacated an arbitral award expressly because of AI "hallucinations" in the reasoning process.

What the Court Decided

The Superior Court found that the award rested, at least in part, on judicial decisions cited by an AI tool that turned out to be fabricated. The winning party discovered the problem and brought the issue before the court, which then invalidated the award rather than letting it stand.

The decision sits at the intersection of two rapidly developing areas of law: the standards for judicial review of arbitral awards and the professional obligations surrounding the use of generative AI in legal practice.

Why It Matters for Practitioners

The case is a wake-up call for the legal profession. AI tools can produce convincing but entirely invented case citations. When those hallucinated authorities make their way into submissions — or, as here, into an award itself — the consequences can be severe: the entire decision may be voided, no matter how much work went into it.

For arbitrators, the message is direct. Verify every authority before relying on it. An arbitrator's failure to confirm that cited cases actually exist can undermine the integrity of the award and expose it to being set aside.

For counsel, the risks are equally significant. Law societies across Canada have issued guidance on generative AI use, and courts are increasingly willing to sanction lawyers who file materials containing fabricated citations. The ARIHQ v. Santé Québec decision extends that scrutiny to the arbitral context.

The Broader Context

The decision arrives amid a wave of AI-related incidents in litigation. Courts in several jurisdictions have encountered fake citations in lawyer filings, drawing fines, reprimands, and dismissed motions. This case moves the issue one step further: the hallucinated authorities were not merely in party submissions but influenced the award's reasoning, prompting judicial intervention at the enforcement stage.

The case also tests the boundaries of supervisory review. Setting aside an award is an exceptional remedy. That the Superior Court took this step on AI-hallucination grounds signals that courts will treat the integrity of the decision-making process — including the authenticity of the legal sources relied upon — as a threshold requirement that arbitration must satisfy.

Practical Consequences

Practitioners should take several lessons from the ruling. First, any legal research produced with AI assistance requires independent verification of every citation before it is filed or relied upon. Second, parties who discover hallucinated authorities in an award have a viable path to challenge it before the courts. Third, arbitral institutions and tribunals may respond by adopting explicit protocols for disclosing and verifying AI-assisted research.

The judgment in ARIHQ v. Santé Québec will likely feature prominently in future debates over AI governance in the legal profession, both in Canada and internationally.