Who Wrote The Award? Artificial Intelligence Decided Awards As The Next Frontier For Challenging Arbitral Awards: Lessons From ARIHQ v Santé Québec And LaPaglia v Valve Corporation

1. Introduction

1.1. The emergence and rapid advancement of generative artificial intelligence (“Generative AI” or “AI”) as a sophisticated tool capable of performing tasks that, until recently, were thought to require exclusively human input has reshaped virtually every sector. Naturally, and as with other sectors, Generative AI has found its way into all facets of the legal sector, arbitration inclusive. This development is hardly surprising as arbitration has historically evolved alongside technological advancement. In fact, one may argue, and rightfully so, that arbitration and technology, although distinct in their origins and functions, are two buildings resting on the same foundation, which is the desire to make human processes simpler and faster. There is therefore little reason, in principle, why Generative AI should be treated differently.

1.2. The nature of arbitration, however, introduces a particular difficulty. Arbitration derives its legitimacy from the parties’ deliberate decision to entrust the resolution of their dispute to a neutral decision-maker possessing the expertise and independent judgment necessary to determine their rights and obligations. This implicit bargain between the parties lies at the heart of every arbitration agreement. The integration of Generative AI consequently raises a fundamental question concerning the legitimacy of the arbitral process. At what point does Generative AI cease to function as an assistive tool and begin to replace the independent judgment that the parties contracted to receive? Put differently, if an arbitrator relies extensively on Generative AI in preparing an award, can it still be said that the award represents the arbitrator’s own decision? Or has the arbitrator impermissibly delegated the adjudicative function entrusted exclusively to him or her?

1.3. These questions have recently come to the fore in Association des ressources intermédiaires d’hébergement du Québec (ARIHQ) v. Santé Québec (2026 QCCS 1360) (the “ARIHQ Case”) decided by the Superior Court of Québec, Canada, and LaPaglia v. Valve Corporation, No. 3:25- cv-00833-RBM-DDL (the “LaPaglia Case”) decided by the United States District Court for the Southern District of California, United States of America.

1.4. This article delves into the emerging intersection between Generative AI and arbitration, interrogating the reasoning of the courts in the ARIHQ and LaPaglia cases and what these developments portend for the future of arbitral decision-making. It further places the Nigerian legal framework under a magnifying glass, examining whether the existing safeguards under the Arbitration and Mediation Act, 2023 (the “AMA 2023”), particularly the grounds for setting aside arbitral awards are sufficiently equipped to address the novel challenges posed by the use of Generative AI by arbitral tribunals.

2. The ARIHQ and LaPaglia Cases

2.1. The ARIHQ Case

2.1.1. In the ARIHQ Case, the parties' relationship was governed by a National Agreement which contained an arbitration clause requiring any disagreement to be notified in writing within 90 days before being referred to arbitration before a sole arbitrator. The dispute was eventually referred to arbitration before a sole arbitrator who, in his final award delivered on August 8, 2025 (the “Award”), upheld the objection that the 90-day contractual time limit had not been complied with and therefore dismissed the claim without considering its merits.

2.1.2. The applicants subsequently sought an annulment of the Award at the Québec Superior Court (the “Quebec Court”). The applicants advanced, principally, two (2) grounds of challenge. They argued, first, that the 90-day contractual time limit for notifying a disagreement was contrary to public policy, on the basis that it impermissibly shortened the three-year statutory limitation period applicable to contractual claims under Québec law, and secondly, that the arbitral procedure agreed by the parties had not been respected because the Award relied on non-existent doctrinal and jurisprudential authorities, which suggested that the Award had been prepared using AI. The first ground was rejected, and the second ground led to the annulment of the Award and gave rise to the issue that is the focus of this discourse.

2.1.3. The applicants identified several judicial and doctrinal authorities cited in the Award which could not be found and argued that their presence indicated that the arbitrator had relied upon Generative AI in preparing his decision. Upon examination, the Quebec Court found that several of the references relied upon by the arbitrator in support of his own reasoning were fictitious. Significantly, these were not merely authorities reproduced in the course of summarising the parties' submissions, they formed part of the arbitrator's own legal reasoning and supported the conclusions reached in the Award. The Quebec Court treated these fictitious authorities as evidence of AI “hallucinations”. These findings led the thought process of the Quebec Court to ultimately annul the Award.

2.2. The LaPaglia Case

2.2.1. The LaPaglia proceedings present a different procedural posture but raise a strikingly similar question. The petitioner, John LaPaglia, sought to vacate an arbitral award delivered on January 7, 2025 in favour of the respondent, Valve Corporation following an arbitration concerning, among other matters, alleged antitrust violations. The petition relied on three (3) principal grounds: (i) that the arbitrator had improperly consolidated the petitioner’s claims with those of other claimants; (ii) that the arbitrator had refused to admit material expert evidence; and (iii) that the arbitrator had outsourced his adjudicative role to AI, which the petitioner argued, under section 10(a)(4) of the Federal Arbitration Act (the “FAA”), amounted to the arbitrator exceeding the authority conferred on him by the parties.

2.2.2. The principal complaint of the petitioner was that the arbitrator had delivered an award that appeared to contain information which had entered the reasoning from somewhere other than the evidence and submissions presented during the arbitration. According to the petitioner, first, during breaks in the arbitration, the arbitrator allegedly disclosed that he had previously used AI to write articles. The petitioner relied on this as evidence that the arbitrator was familiar with, and willing to use, AI as a writing tool.

2.2.3. Second, the petitioner pointed to the speed with which the award was produced. The arbitration had lasted ten (10) days and generated approximately 2,000 pages of transcript. The final post-hearing brief was submitted on December 23, 2024, yet the arbitrator issued a 29-page award only 15 days later, on January 7, 2025. The timing was particularly significant because, according to the petitioner, the arbitrator had told the parties during breaks in the proceedings that he wanted to complete the decision before a scheduled trip. The petitioner argued that the combination of the bulky evidence, the length of the hearing and the unusually short period between the final submission and the award supported an inference that AI had been used to produce the decision.

2.2.4. Third, the petitioner alleged that the award contained factual assertions which he described as “tell-tale signs of AI generation” that were not presented at the arbitration and were unsupported by the evidentiary record. For example, the petition identified statements concerning Roblox, Sony, Microsoft and competition in the Chinese gaming market which, according to the petitioner, had neither been introduced into evidence nor supported by citations.

2.2.5. The respondent, however, did not engage the Court on the substantive question of the alleged use of AI. Instead, the Respondent challenged the Court's jurisdiction, arguing that the FAA itself does not confer federal jurisdiction and that the petitioner had failed to establish an independent basis for federal jurisdiction. The Court accepted this argument and held that the FAA did not itself supply federal jurisdiction and that the petition did not otherwise establish the necessary jurisdictional basis. The Petition was accordingly dismissed on December 9, 2025.

3. AI and the Personal Exercise of the Arbitral Function

3.1. As earlier emphasised, arbitration has always evolved alongside technological advancement and it is undeniable that technology has become an indispensable feature of modern arbitral practice. For example, the emergence of online dispute resolution, particularly in response to the growth of e-commerce, saw the use of online negotiation, mediation and arbitration before virtual hearings started to gain entry into litigation. Similarly, electronic bundles and document management systems have largely replaced the physical handling of arbitral records. Accordingly, it may seem unreasonable to posit that AI should not be utilised in the preparation of awards, at all. The more critical issue for discussion is the degree of utilisation and whether there exists a point at which the use of AI ceases to be merely assistive and instead amounts to an impermissible delegation of the arbitrator’s adjudicative function.

3.2. The starting point for this inquiry is the consensual nature of arbitration itself. Unlike litigation, where the authority of the courts is derived from the State, arbitrators derive their authority from the arbitration agreement executed between the parties, which is in its most granular form, an agreement to submit a dispute to the judgment of a particular decision-maker chosen by the parties by virtue of the decision-maker’s expertise. The manner in which arbitrators are selected is instructive in this regard as the appointment of an arbitrator is rarely an incidental exercise. Parties and their counsel undertake considerable research into the prospective arbitrator’s professional background, areas of expertise, prior experience, publications, previous appointments and, where applicable, the composition of panels on which the arbitrator has previously sat. These considerations are relevant precisely because the parties are selecting the individual whose judgment they consider best suited to resolving the particular dispute before them. The arbitrator, in turn, assumes that mandate personally and is ordinarily required, upon appointment, to make the requisite declarations of independence and impartiality. The arbitrator’s mandate is accordingly a personal one as it is the arbitrator’s expertise that the parties bargain for and entrust with the resolution of their dispute.

3.3. This personal nature of the arbitral mandate is, however, not inconsistent with the use of assistants. It is standard practice in international arbitration for arbitral tribunals to receive assistance in the administration and preparation of arbitral proceedings. Tribunal secretaries, for example, may assist with procedural administration, document organisation, legal research, proof-reading and the preparation of drafts, including factual portions of awards, provided that such material remains subject to the tribunal's supervision and is subsequently reviewed by the tribunal itself. This practice of permitting assistance has never been understood as permitting the delegation of the arbitrator's adjudicative function. To buttress this point, reputable arbitral institutions such as the International Chamber of Commerce and the London Court of International Arbitration recognise the role of tribunal secretaries in their respective arbitration frameworks, but expressly provide that the functions performed by a secretary must not amount to a delegation of the tribunal's decision-making authority.1

3.4. The same principle can reasonably be extended to Generative AI. If an arbitrator cannot delegate substantive decision-making to a human assistant without the parties’ consent, it is difficult to see why the same substantive decision-making could properly be delegated to an algorithm. Once an arbitrator permits Generative AI to undertake functions such as the evaluation of witness credibility, the weighing of evidence, the determination of disputed facts, the interpretation and application of legal principles and the formulation of the reasoning supporting an award which all constitute aspects of the personal mandate entrusted to the arbitrator, the concern is no longer one of technological assistance but of impermissible delegation.

3.5. In the ARIHQ Case, the gravamen of the Quebec Court’s decision was not that the arbitrator had used AI, but that the circumstances surrounding its use undermined confidence that the Award could be regarded as the product of the arbitrator’s own reasoning and intellectual rigour. The fabricated authorities suggested that the arbitrator had failed to exercise the independent judgment, diligence and legal expertise for which the parties had acceded to his appointment. The LaPaglia Case exposes the same concern from a different angle. The petitioner did not merely object to the use of Generative AI as a drafting aid but alleged that the AI had dabbled into the realm of adjudication by supplying facts and reasoning which the arbitrator adopted in reaching his decision. Whether that allegation would ultimately establish impermissible delegation was never determined. Nevertheless, the petition illustrates the direction in which future challenges are likely to develop.

4. AI-assisted Arbitration within the Nigerian Context

4.1. The AMA 2023 contains no express provision regulating AI-assisted arbitral proceedings, and, to the best of our knowledge, there is presently no Nigerian judicial decision determining whether the use of Generative AI in preparing an arbitral award constitutes a ground for setting aside the award. However, can it be said that the AMA 2023, in its present form, is sufficiently adaptable to accommodate the novel challenges that AI-assisted arbitration may present, or whether the emergence of these challenges will ultimately expose gaps in the existing statutory framework?

4.2. To be sure, section 55 of the AMA 2023 provides that recourse against an arbitral award may be made only by an application to set aside and, importantly, provides that such an application cannot be founded on an error on the face of the award or any ground other than those expressly stated in subsection (3). This closed wording means that a Nigerian court cannot simply create a free-standing ground of “AI misuse” within the context of “error on the face of the award” merely because AI was involved. Nonetheless, a provision worthy of note is section 55(3)(a)(vi), which permits the court to set aside an award where “the arbitral procedure, was not in accordance with the agreement of the parties”. This ground appears to provide a particularly useful route for an AI-based challenge.

4.3. As established earlier, the parties’ agreement to arbitration necessarily involves an agreement to submit the dispute to the decision of an arbitral tribunal. Where an arbitrator merely uses AI as a tool to organise documents or check grammar, it would be difficult to argue that the agreed procedure has been breached. Where, however, a party is able to demonstrate, as in the ARIHQ Case, that the arbitrator delegated substantive fact-finding or legal reasoning to AI, the argument becomes materially stronger. The parties agreed to have their dispute determined by the appointed arbitrator, not an AI system. Accordingly, where an arbitrator delegates a substantive part of the adjudicative function to AI, a challenge may lie under section 55(3)(a)(vi) of the AMA 2023 on the ground that the agreed arbitral procedure has not been followed.

4.4. Another ground that may be relevant in challenging an AI-decided award is the public policy ground under section 55(3)(b)(ii) of the AMA 2023, which allows an award to be set aside when it is against the public policy of Nigeria. An argument may be anchored, in the first instance, on section 91 of the AMA 2023, which defines an “arbitrator” as a “person to whom a reference is made for determination” and an “arbitral tribunal” as “a sole arbitrator or a panel of arbitrators”. The use of the term “person” suggests that the law contemplates a human decision-maker entrusted with the adjudicative function. Accordingly, where an arbitral tribunal delegates the substantive decision-making function to Generative AI, it may be argued that the arbitral tribunal has, in substance, transferred the authority to determine the dispute to a machine that cannot constitute an “arbitrator” within the meaning of section 91 of the AMA 2023. This would be inconsistent with the fundamental objective of the AMA 2023, which entrusts the resolution of the parties’ dispute to an arbitral tribunal.2 On that basis, an award resulting from such delegation may be characterised as being contrary to the public policy of Nigeria within the meaning of section 55(3)(b)(ii) of the AMA 2023.

5. Conclusion

5.1. The arrival of Generative AI in arbitration is no longer something to be debated at the margins. It has found its way into all the crevices of the practice. That should not, in itself, trouble us as arbitration has never been hostile to technology. Indeed, much of what now seems ordinary in arbitration, from electronic bundles to virtual hearings, was once new.

5.2. There must, however, be a line. The cases considered in this article suggest that the line should not be drawn at the point at which an arbitrator first uses AI, but should be drawn at the point at which the arbitrator ceases to do the job for which the parties appointed him or her. The ARIHQ Case is instructive because the fictitious authorities and academic references raised a more fundamental concern of whether the reasoning in the Award was in fact the arbitrator’s own reasoning. The LaPaglia Case presents the other side of the difficulty, in that proof of such delegation may be difficult where the allegation rests on circumstantial evidence. A polished award, an unusual style or even an error cannot, by itself, establish that an arbitrator has delegated the decision-making function. The question must ultimately be whether there is sufficient evidence that AI has supplanted, rather than assisted, the arbitrator’s judgment.

5.3. With respect to the Nigerian context, the AMA 2023 says nothing expressly about AI-assisted arbitration, and, to the best of our knowledge, our courts have not yet had to decide whether the use of AI in preparing or writing an award may justify setting an award. However, section 55(3)(a)(vi) of the AMA 2023, which allows an arbitral award to be set aside when the arbitral procedure agreed by the parties was not followed, and section 55(3)(b)(ii) of the AMA 2023, which allows an arbitral award to be set aside when it is against the public policy of Nigeria, offer a pathway to challenge AI-decided awards. It would be interesting to see how a Nigerian court would decide such a challenge.

5.4. It goes without saying that the increasing demand and use of Generative AI clamours for some form of regulation on reliance by arbitral tribunals on AI in deciding arbitrations. It has become imperative for the legislature to address this growing concern by amending the AMA 2023 to put some safeguards in place and/or introduce provisions that make delegation of substantive decision-making to AI an express ground for setting aside an arbitral award. However, any such reform must be carefully and comprehensively considered, with appropriate thresholds established to avoid an overly broad prohibition of the use of AI or abuse of the setting-aside mechanism.

5.5. Perhaps, this is ultimately the right way to think about the problem. AI is not the first technology to be utilised in arbitration, and it will not be the last. The future of AI-assisted arbitration should accordingly not be characterised as a choice between technological progress and the preservation of arbitral integrity. The two are capable of co-existing. Arbitration has always embraced tools that make the resolution of disputes faster, more efficient and more accessible and AI is simply the latest iteration of that development. The task of the law is not to prevent arbitrators from using the technology, but to ensure that the technology remains subordinate to the mandate entrusted to the arbitrator by the parties. AI may assist the arbitrator or even augment the arbitrator’s expertise, but it cannot become the arbitrator.

Footnotes

^{1. Note to Parties and Arbitral Tribunals on the Conduct of ICC Arbitration (2026), para. 202; London Court of International Arbitration Rules (2020), art. 14.}

^{2. The AMA 2023, s. 1(1).}

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.