Awards challenged over AI use: what courts have decided so far

What the Québec court decided

In ARIHQ c. Santé Québec, 2026 QCCS 1360, two claimants applied to annul a sole arbitrator’s award of 8 August 2025. The award had dismissed a claim for unpaid remuneration of about C$1.2 million because a contractual notice was late. One ground for annulment was that the arbitration procedure had not been observed, under article 646 of Québec’s Code of Civil Procedure, because the award relied on doctrine and case law that do not exist ([1] to [3], [23]).^1,2

The court rejected a separate public order ground ([55], [64] to [66]) but upheld the procedure ground. The court said that parties who choose an arbitrator are entitled to a decision by that arbitrator, so the decision cannot be delegated to anyone else ([75], [81] to [86]). The rule does not stop the use of researchers, clerks or help with translation or citations. But anyone else’s part in the drafting must not undermine the integrity of the process, and responsibility for the drafting must stay with the arbitrator, so no one could think a third party had taken over the writing of the reasons ([87]).¹

On the evidence, the court found that every doctrinal and case-law reference the award relied on either did not exist or did not support the point cited ([102] to [113]). It concluded that the arbitrator’s authority had been delegated ([114] to [116]), annulled the award and ordered the parties to choose a new arbitrator within 60 days ([122] to [124]).¹

On AI itself, the court accepted the defendant’s point that AI use is not one of the grounds listed in article 646, but said that did not settle the question ([69]). It said nothing is inherently wrong with using a reliable AI tool, but that lawyers and parties must check what they file ([94]), and that the pitfalls are greater still for decision-makers ([95]).¹

The court also set limits on its ruling. It said that not every award that cites wrong authorities or uses AI as a drafting tool should be annulled. Minimal use, or use on a minor issue, may lead to a different result ([117], [118]).¹

What the US court did not decide

In LaPaglia v. Valve Corporation, a consumer asked a federal court in California to vacate an AAA award under section 10 of the Federal Arbitration Act. One of the grounds was that the arbitrator had exceeded their powers by outsourcing the decision to AI. The petition relied on a remark the arbitrator was said to have made about using ChatGPT for an unrelated article, and on the speed and content of the award.^3,4

On 9 December 2025 the court dismissed the amended petition for lack of subject-matter jurisdiction. The petition did not arise under federal law, and an award of zero dollars did not meet the amount in controversy for diversity jurisdiction (Order, pp. 7 to 12). The court never reached the AI argument, so the case decides nothing about whether AI use can justify vacating an award.³

The cases side by side

Context from England: false citations in court filings

R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank, [2025] EWHC 1383 (Admin), is not an arbitration case. The Divisional Court dealt with two unrelated cases in which lawyers put citations before the court that did not exist, 18 of 45 citations in one of them. It held that lawyers who use AI for legal research must check it against authoritative sources before relying on it. Putting false citations before the court can lead to sanctions, from a public admonition or costs orders up to referral to a regulator, contempt proceedings or referral to the police ([7] to [9], [23] to [31]).⁵

It is included because it shows how courts treat unchecked AI output in general. It says nothing about arbitrators or awards.

What the institutions’ rules say about the same risk

The published guidance points the same way as the Québec judgment. SVAMC’s Guideline 6, the SCC guide and the VIAC note each say that an arbitrator must not hand the decision to an AI tool, and Ciarb’s guideline says an arbitrator’s own use of AI should not replace their judgement. The AI rules tracker sets out what each text says, including on disclosing AI use.^6,7,8,9

What is not settled yet

• Whether the Québec judgment will be challenged further. When we checked CanLII on 21 October 2026, it listed no later decision in the case.
• How much AI use is too much. The Québec court left open how it would treat minimal use, or use on a minor issue ([117], [118]).
• How the analysis would run under an appeal or judicial review standard, which the Québec court also left open ([119]).
• Whether a US court would vacate an award on this ground. The one US challenge we found ended without a ruling on it.^1,3