The perils of the misuse of artificial intelligence (“AI“) in the legal services industry have mainly affected implicated legal practitioners around the world. These AI hazards are now snowballing to entangle Judges and arbitrators. On 22 April 2026 –and in what is likely to be the first of its kind– in Association des ressources intermédiaires d’hébergement du Québec (ARIHQ) c. Santé Québec 2026 QCCS 1360 (“ARIHQ Judgement“), the Canadian Quebec Superior Court annulled an arbitration award because the arbitrator delegated his decision-making powers to an AI tool.

Internationally, the script in these AI developments is similar: a legal practitioner (and now Judges and arbitrators) use AI tools to conduct legal research, AI tools, in their inventiveness, use nonexistent sources to support legal arguments or conclusions, no one checks these references immediately and when someone does ultimately check, the sources quoted by AI tools are found to be nonexistent. The reaction of the international community has been diverse. In South Africa, there have been public admonishments by Judges and referral of legal practitioners to the Legal Practice Council for investigations and possible disciplinary proceedings. In the United Kingdom, there have been reprimands by Judges and judicial pronouncements that legal practitioners who place false material before courts may be prosecuted criminally for perversion of justice or contempt of court. In the United States, the judicial censure has escalated from reprimands to fines.

WHAT HAPPENED IN THE ARIHQ JUDGEMENT?

On 7 June 2018, the Minister of Health and Social Services (“MSSS“) and the Quebec Association of Intermediate Housing Resources (“ARIHQ“) concluded an agreement referred to as the National Agreement (“Agreement“). In 2021, a dispute arose relating to the Agreement. That dispute involved payment of $ 1,225,000. Notwithstanding various settlement discussions, no amicable resolution was found. On 1 August 2024, and to escalate the dispute resolution mechanisms in the Agreement, ARIHQ issued a notice of disagreement (“Notice of Disagreement“). On 28 August 2024, the Integrated University Health and Social Services Centre of South-Central Montreal (“CCSMTL”) objected to the Notice of Disagreement. It argued that in terms of the Agreement, the Notice of Disagreement had to be issued within 90 (ninety) days from the date on which the referring party became aware of the dispute and that this period could only be extended by written agreement between the parties.

On 13 September 2024, ARIHQ referred the dispute to arbitration. In response, CCSMTL raised a preliminary issue that the dispute resolution referral should be dismissed on the basis that the Notice of Disagreement was filed out of time.

On 7 July 2025, the arbitration was heard. On 8 August 2025, the arbitrator issued his award dismissing the dispute referral on the basis that the Notice of Disagreement was filed out of time.

In November 2025, ARIHQ initiated court proceedings for the annulment of the arbitration award.

ARIHQ advanced two arguments for setting aside the award. The first is irrelevant for the purposes of this note. The second was as follows –

“The decision [arbitration award] does not comply with the arbitration procedure, since the Arbitrator delegated part of his decision-making power to a generative artificial intelligence tool (“AI“).” [at paragraph [41] Emphasis added]

WHAT WERE THE ARGUMENTS RAISED BY THE CCSMTL IN DEFENSE

The CCSMTL raised the following arguments in opposition to the request for the award annulment –

• that the use of AI is not a ground for invalidation listed in Article 646 of the Canadian Code of Civil; and
• that in the absence of expert evidence, the Court could not conclude that AI was used or that the cited authorities did not exist.

WHAT DID THE COURT SAY?

Justice Sheehan quoted about eight references in the arbitration award to sources of law that were hallucinated by AI and were therefore either nonexistent or cross-referred to irrelevant laws. He then set aside the arbitration award and referred the matter to a new arbitrator to be appointed by the parties. He said the following –

“[113] The references mentioned above are central to the Arbitrator’s reasoning. They constitute the only doctrinal or jurisprudential references used as legal support for the Award. Other jurisprudential references are included in sections of the Award that summarize the parties’ positions.

[114] The preponderance of evidence therefore leads to the conclusion that the Arbitrator’s authority was delegated and that he abdicated his role in reviewing the result.

[115] This conclusion is inevitable given that all the doctrinal and jurisprudential references on which the Arbitrator relies are non-existent and ” hallucinatory“.

[116] For this reason, the sentence must be overturned.

[117] This conclusion does not imply that every judgment that cites erroneous references or uses artificial intelligence as a drafting tool should suffer the same fate. For example, it is conceivable that there will be cases where the use of artificial intelligence will have been minimal or will address a less significant issue.

[118] In such cases, the weighting of the nature of the breach in relation to the arbitration proceedings initiated, the determination of the infringement of the integrity of the proceedings, and the assessment of the impact of the breach on the award are crucial could lead to a different result.” [Emphasis added]

In regard to the argument by the CCSMTL that under the Canadian laws the use of AI tools is not a ground for invalidation of arbitration awards, Justice Sheehan said the following –

“[71] …even though the legislator did not specify each of the procedural violations that could justify annulment, it should be understood that any ” significant violation of the procedure agreed upon by the parties ” that calls into question ” the integrity of the arbitration process ” can lead to such an annulment.  Conversely, ” a formal and insignificant procedural irregularity, which could not have caused any serious harm or injustice and could not have had the slightest effect on the content of the sentence or the outcome of the dispute, will not give rise to censure.” [Emphasis added]

In response to the argument that in the absence of expert evidence the Court could not conclude that the arbitrator used AI tools in decision-making, Justice Sheehan said –

• the burden of proof in civil matters rests on the balance of probabilities; and
• the fact that the decisions referred to by the arbitrator did not exist had been established since each of the neutral references cited by the arbitrator led to other decisions that have no connection to the subject matter in support of which they were cited.

In relation to the delegation of decision-making powers, the Justice said the following –

“[86] The primacy given to the autonomy of the parties’ will in choosing the arbitrator, the importance of drafting the reasons to ensure an informed decision, the arbitrator’s duty to maintain the secrecy of the deliberations, and the imperative to maintain public confidence in the arbitration process justify that decisions be drafted by the arbitrator chosen by the parties without delegation to third parties.

[87] The rule is not intended to prohibit the use of researchers, clerks, translation or citation assistance, or even reference to notes and authorities prepared by lawyers. The rules of natural justice do not prevent decision-makers from consulting colleagues. Furthermore, the involvement of third parties in the drafting process must not compromise the integrity of the process and must maintain the decision-maker’s responsibility for the drafting in order to avoid any interpretation that a third party has taken control or been delegated the task of drafting the reasons. The maxim ” he who decides must hear ” is a consequence of the rule audi alteram partem ” insofar as a litigant is only truly ‘ heard ‘ if he is heard by the one who will decide his case“.” [Emphasis added]

Further, Justice Sheehan identified the following five risks of using AI tools in decision-making –

• first, the use of AI may lead to the creation of false references or hallucinations that are extremely deceptive and difficult to detect;
• second, the rigidity of computer programs such as AI toolsis incompatible with discretionary decisions and are unable to factor in decision-making community values and any other relevant contextual circumstances;
• third, AI tools may introduce bias in decision-making because they operate based on the specific data used in each platform;
• fourth, the feeding of information to AI systems may violate the obligation of an arbitrator to keep arbitration deliberations secret; and
• fifth, the misuse of AI may negatively affect public confidence in the justice system; in this regard, he said the following –

“Our justice system only works if people trust it. The use of artificial intelligence in procedures that may then influence judgments can breed cynicism toward the legal profession and the justice system. Justice requires that decisions be perceived as fair, humane, and reasoned. If the public believes that a disembodied or opaque machine has made a decision, rather than a human being, trust in the justice system could be shaken.” [Emphasis added]

WHAT DOES THIS MEAN IN THE SOUTH AFRICAN CONTEXT?

In South Africa, section 33(1) of the Arbitration Act, No. 42 of 1965, as amended, sets out the following four grounds on which a Court may set aside an arbitration award –

• where the arbitrator has “misconducted himself in relation to his duties as arbitrator” (section 33(1)(a));
• where the arbitrator has committed “gross irregularity in the conduct of the arbitration proceedings” (section 33(1)(b));
• where the arbitrator “has exceeded his powers” (section 33(1)(b)); and/or
• when “an ward has been improperly obtained” (section 33(1)(c)).

Therefore, if an arbitrator delegates his or her decision-making powers to AI as in the ARIHQ Judgement, the affected award may be set aside on the basis that the arbitrator misconducted himself or herself. In Cheg Trading (Pty) Ltd and Another v Emfuleni Estate Home Owners Association and Another (https://www.saflii.org/za/cases/ZAGPPHC/2016/212.html), Acting Judge Olivier described misconduct in the context of section 33(1)(a) of the Arbitration Act as follows

“[29] It seems to me that if misconduct is given its ordinary meaning, the arbitrator should have engaged in some sort of wrongful or improper conduct for it to be challenged. Moral turpitude or mala fides should be established. A bona fide mistake cannot be misconduct… Although an arbitrator is under no obligation to observe the strict rules of evidence and procedure that would ordinarily apply in a court of law, he must at least ensure that the parties are given a fair and just hearing, informed by the principles of natural justice. He must discharge his duties impartially and honestly…” [Emphasis added]

It is now internationally accepted that those who use AI tools to conduct legal research have a duty to ensure the veracity of the outcomes of the research and that failure to do so may constitute unprofessional conduct. Therefore, it should follow that the delegation of decision-making powers by an arbitrator, which result in an award that fundamentally relies on nonexistent laws, constitute improper conduct and therefore misconduct in terms of section 33(1)(a) of the Arbitration Act.

Although gross irregularity generally relates to the conduct of the arbitrator during the arbitration proceedings, there are various grounds on which it may be argued that the delegation of decision-making powers constitutes gross irregularity under section 33(1)(b) of the Arbitration Act.

The following are practical lessons for the arbitration stakeholders from the ARIHQ Judgement in South Africa –

• the use of AI tools in arbitration proceedings must be addressed in pre-arbitration meetings and commercial agreements recording resolution of disputes through arbitration;
• since AI tools are known to be useful and are being used, arbitrators who use these tools must meticulously verify the research output;
• arbitration bodies must develop and adopt AI guidelines to assist those participating in arbitration proceedings, including arbitrators; and
• legal practitioners must scrutinize arbitration awards for potential references to hallucinated precedents and institution of review proceedings to set aside the affected awards.

On 15 May 2025, the Association of Arbitrators (Southern Africa) NPC adopted a document titled “Guidelines on the Use of Artificial Intelligence (“AI”) in Arbitrations and Adjudications” (https://www.arbitrators.co.za/resources/ai-guidelines/ (accessed on 29/05/2026)). In regard to decision-making by arbitrators, these guidelines provide as follows –

“14.1 In decision-making and award process, the use of AI must be strictly controlled by Tribunals. Delegation of decision-making authority should be strictly prohibited; this includes delegating to AI Tools. Any AI generated information must undergo independent verification before inclusion in decisions. Biases inherent in AT Tools must be systematically identified and mitigated. Tribunals are accountable for all aspects of awards, irrespective of the extent of AI utilisation. Comprehensive documentation should be maintained, including records of decision-making processes and AI-related activities.

14.2 Tribunals should ensure, as far as possible, that the award will withstand challenge. To this end, consideration should be given to the extent to which AI has been used during the preparation of the award, the extent to which disclosure within the award of its use should be made, and any other aspects to demonstrate that the award is essentially the work of the Tribunal, assisted, where warranted, by the use of AI Tools. ” [Emphasis added]

Insofar as South African Courts are concerned, the ARIHQ Judgement demonstrates an urgent need for the finalisation of AI guidelines for South African Courts. On 18 May 2026, it was reported that the High Court of South African, Gauteng Division, Johannesburg has adjudicated an appeal based on the alleged improper use of AI tools by an Acting Judge (“Acting Judge allegedly used AI to prepare ruling” https://legalbrief.co.za/diary/legalbrief-africa-new/story/acting-judge-accused-of-using-ai-to-prepare-ruling-4/).