A court has killed a Montreal arbitrator's decision after the ruling was found to rest on artificial-intelligence hallucinations — legal authorities that the technology simply invented.

Law360 reports that the arbitration award was set aside because the arbitrator relied on AI-generated citations that do not exist. The decision adds Montreal to the growing list of jurisdictions where courts have scrapped adjudicative outcomes tainted by fabricated case law produced by generative AI tools.

What happened?

The arbitrator issued a decision in a Montreal arbitration. A party subsequently challenged the award. The reviewing court found that the reasoning depended on hallucinated legal material — cases or authorities generated by an AI system that turned out to be nonexistent.

The court killed the decision as a result. The award cannot stand where its legal foundation includes authorities that were never real.

Why does this matter for practitioners?

The ruling is a direct warning to arbitrators and litigators who use generative AI in drafting awards, briefs or opinions. Tools such as large language models can produce convincing citations with full case names and quotes — and the underlying cases may not exist.

For arbitration counsel, the practical consequences are concrete:

• Verify every citation in an award or submission before signing or relying on it.
• Expect opposing parties to check authorities cited against you; fabricated cases expose filings to challenge.
• Arbitral awards, once tainted by hallucinated reasoning, risk being set aside — destroying the finality that arbitration exists to provide.
• Disclosure and verification protocols for AI use are becoming a baseline professional expectation.

A growing pattern

Courts across multiple jurisdictions have now sanctioned lawyers or overturned outcomes where AI-fabricated cases appeared in filings. The Montreal arbitration decision extends that trend from court pleadings into the arbitral context, where a neutral decision-maker — not just a party's counsel — relied on hallucinated authorities.

That distinction sharpens the stakes. When an arbitrator's own reasoning rests on invented law, the award itself becomes vulnerable, and the parties lose the benefit of their bargain for a binding, durable resolution.

For arbitrators, the lesson is procedural discipline: treat AI output as an unverified draft, confirm every authority against primary sources, and document that verification. For counsel, the lesson is vigilance — including toward the tribunal.

The decision stands as one of the clearest examples yet of AI hallucinations directly destroying an adjudicative outcome in Canada's arbitral system.