Artificial intelligence (“AI”) is increasingly being used in international arbitration for legal research, document review, translation, and drafting. However, its use becomes difficult to assess when an arbitrator turns to AI not merely to process information, but to obtain a view on a legal issue that the tribunal must decide.
There is presently no generally applicable rule across international arbitration prohibiting an arbitrator from putting legal questions to AI. Indeed, in February 2026, UNCITRAL devoted part of a colloquium specifically to “Arbitrators and the Use of AI – Impact on Recognition and Enforcement”, reflecting the fact that the issue remains under consideration internationally.[1]
The emerging guidance nevertheless points towards an important distinction. AI may assist an arbitrator, including with legal research and the processing of information, but the arbitrator must retain responsibility for the legal analysis and the ultimate decision.
Asking AI a Legal Question Is Not Necessarily Delegating the Decision
Arbitrators have long relied on research tools and assistants. They consult legal databases, textbooks, commentaries and previous awards. Tribunal secretaries may review authorities and summarise submissions. An arbitrator does not cease to decide a case personally merely because another source helps identify relevant legal material.
The same reasoning may apply to certain uses of AI. Asking an AI system to locate authorities concerning the interpretation of limitation clauses, for example, is functionally closer to legal research than adjudication. Asking the system to identify possible arguments against a tentative interpretation may likewise assist the arbitrator’s analysis, provided that the arbitrator assesses those arguments independently.
The position becomes more difficult where the prompt itself asks the AI system to perform the adjudicative exercise. Questions such as “Which party has the stronger argument?”, “Is the claimant entitled to damages?” or “How should the tribunal decide this jurisdictional objection?” ask the system to evaluate competing positions and reach a conclusion.
The relevant questions, therefore, are what task AI is asked to perform and how its output is used.
Emerging Guidelines Draw the Line at Decision-Making
The 2025 Chartered Institute of Arbitrators Guideline on the Use of AI in Arbitration (“Ciarb AI Guideline”) expressly recognises that arbitrators may use AI tools to enhance both efficiency and the quality of decision-making.[2] It nevertheless provides that arbitrators should not relinquish their decision-making powers to AI and must ensure that their judgment remains independent.[3]
Significantly, the Ciarb AI Guideline advises tribunals to avoid delegating tasks such as “legal analysis, research and interpretation of facts and law, or application of the law to the facts” where the use of AI could influence procedural or substantive decisions.[4] It also requires arbitrators to independently verify information obtained through AI and ultimately to assume responsibility for every aspect of the award.[5]
The Silicon Valley Arbitration & Mediation Center Guidelines on the Use of Artificial Intelligence in Arbitration (“SVAMC Guidelines”) adopt a similar approach. Guideline 6 provides that an arbitrator may not delegate any part of the arbitrator’s personal mandate to an AI tool and specifically states that AI must not replace the arbitrator’s independent analysis of the facts, law and evidence.[6]
Advisory Opinion No. 28 of the National Academy of Arbitrators takes a comparable approach, although it was issued in the context of labour-management arbitration. It characterises AI as a form of “virtual assistant” and distinguishes between the use of AI as an aid and the delegation of a decision-making function.[7]
Although expressed differently, these instruments draw the same line: AI may assist the tribunal, but it cannot replace the tribunal’s judgment.
The Analogy with Tribunal Secretaries
The existing rules governing tribunal secretaries provide a useful analogy.
Article 14.8 of the 2020 LCIA Arbitration Rules permits tribunals to obtain assistance from tribunal secretaries but states expressly that “[u]nder no circumstances” may the tribunal delegate its decision-making function to the secretary.[8] The LCIA Guidance Note nevertheless contemplates secretaries performing potentially substantive tasks, including preparing first drafts of awards and procedural orders, subject to party approval, the tribunal’s supervision and its continued responsibility for the decision.[9]
The 2026 ICC Rules and the ICC Note to Parties and Arbitral Tribunals provide that a tribunal secretary works under the tribunal’s direction and control and may not exercise any decision-making authority.[10] The ICC Note further states that tasks entrusted to a secretary do not release the tribunal from its duty personally to review the file and draft its decisions.[11]
In practical terms, some uses of AI therefore resemble tasks traditionally performed by tribunal secretaries. The analogy is not exact, however. A tribunal secretary is an identified individual whose role can be defined and whose obligations of independence and confidentiality are addressed.[12] By contrast, an AI system is not subject to equivalent personal obligations, and the arbitrator may have limited visibility into its sources, processing or retention of information.
These differences become more significant where AI moves beyond retrieving or organising information and begins to analyse the parties’ positions or formulate legal conclusions.
What If AI Identifies an Argument the Parties Never Made?
A separate issue arises where AI identifies a legal argument or authority that neither party has raised.
Many legal systems permit arbitrators, to differing degrees, to conduct their own legal research or apply the law independently under the principle of iura novit arbiter.[13] That does not necessarily mean that a tribunal may decide a case on a new and dispositive legal theory without hearing the parties.
Article 18 of the UNCITRAL Model Law requires that the parties be treated equally and that each be given a full opportunity to present its case.[14] The New York Convention likewise permits refusal of recognition and enforcement where a party was unable to present its case.[15]
The SVAMC Guidelines address this risk by stating that an arbitrator should not rely upon AI-generated information outside the record without appropriate disclosure and, where practicable, an opportunity for the parties to comment.[16]
Thus, an arbitrator might ask AI to identify possible alternative legal analyses but, if the AI identifies a novel proposition that becomes material to the outcome, the safer course is to put the relevant issue or authority to the parties rather than silently incorporate it into the award.
Issues of Confidentiality and Accuracy
International arbitration often involves confidential or commercially sensitive material, including pleadings, witness statements, contracts, and expert reports. Uploading those materials into an inadequately secured AI system can create risks concerning data retention, access, use for training and disclosure.
The Ciarb AI Guideline therefore stresses that third-party AI systems may create significant confidentiality and cybersecurity risks and recommends examining how information is stored, protected and used before confidential material is submitted.[17] Advisory Opinion No. 28 similarly warns that a provider may retain prompts, case materials and AI outputs.[18]
Apart from concerns of confidentiality, generative AI can produce false authorities, incorrect propositions of law or highly persuasive explanations that conceal significant errors.[19] An arbitrator who uses AI-generated research therefore remains responsible for checking the underlying authority rather than treating the AI response itself as a legal source.[20]
This is particularly important when asking AI for its “opinion”. A plausible answer may reinforce an initial view even where the underlying legal reasoning is weak. Independent verification therefore requires more than checking whether the cited cases exist. The arbitrator must independently examine the relevant law and reach their own conclusion.
Disclosure, Party Consent and the Extent of AI Use
The emerging guidance also does not adopt a uniform position on disclosure. The Ciarb AI Guideline encourages arbitrators to consult the parties before using an AI tool and to refrain from using the specified tool if the parties disagree.[21] On the other hand, the SVAMC Guidelines do not require disclosure of AI use as a general matter, although they require prior disclosure where an arbitrator proposes to rely on AI-generated information outside the record.[22] In the more specialised context of labour arbitration, Advisory Opinion No. 28 requires disclosure where generative AI is used, or is intended to be used, for substantive case management, including research, summarising evidence or drafting an award. It also requires the parties’ mutual consent where AI assists in any decision-making function.[23]
The appropriate approach therefore depends on the task performed by AI. Routine or incidental uses do not raise the same concerns as asking AI to evaluate evidence, formulate legal reasoning or prepare substantive parts of an award.
Developments in institutional arbitration illustrate the significance of party consent. The American Arbitration Association (“AAA”), for example, now offers an opt-in AI-led arbitration procedure for eligible two-party disputes. Under its AI Led Arbitration Rules, the AI Arbitrator analyses the parties’ submissions and prepares a draft award with brief reasoning. A human arbitrator then reviews the draft and may approve, edit or rewrite it before finalising and signing the award.[24]
The AAA procedure is based on the parties’ express agreement to a defined role for AI and therefore differs from an arbitrator independently using a general-purpose AI tool. While party consent may permit more extensive AI involvement, it cannot override mandatory law at the seat or basic requirements of procedural fairness.
Could Asking AI for Its View Endanger the Award?
The consequences of improper AI use have now begun to arise before national courts.
As previously discussed by Aceris Law, in LaPaglia v. Valve Corporation, a party sought to vacate an arbitral award on the allegation that the arbitrator had “outsourced his adjudicative role” to AI.[25] The United States District Court for the Southern District of California dismissed the amended petition in December 2025 for lack of subject-matter jurisdiction and therefore did not determine the merits of that allegation.[26]
A more direct judicial response emerged in Québec in April 2026. The Québec Superior Court set aside (unofficial English translation here) an award after finding that non-existent authorities and doctrine were central to the arbitrator’s reasoning and that the circumstances demonstrated uncontrolled use of AI and an improper delegation of part of the adjudicative task.[27]
The Court did not treat the mere use of AI as a ground for annulment.[28] The problem was the arbitrator’s reliance on fabricated material, failure to verify the result and the resulting departure from the procedure agreed by the parties.
Depending on the applicable arbitration law and the circumstances, excessive or unverified AI involvement could give rise to challenges concerning a party’s inability to present its case, a departure from the agreed arbitral procedure or conflict with public policy.[29] At the enforcement stage, analogous grounds appear in the New York Convention.[30]
The relevant inquiry therefore is whether the use of AI displaced the tribunal’s independent judgment or affected the fairness of the proceedings.
Conclusion
Can an arbitrator ask AI for its view on a legal issue?
There is no general rule prohibiting an arbitrator from using AI to identify authorities, expose counterarguments or test a legal analysis. The arbitrator must, however, evaluate the output independently and remain responsible for the reasoning ultimately adopted.
The position becomes more difficult where AI is asked to weigh the parties’ competing cases, formulate the tribunal’s reasoning or recommend the outcome. At that point, the issue is whether AI remains a tool assisting the arbitrator or has begun to perform part of the adjudicative function entrusted to the arbitrator.
The practical dividing line is therefore not whether an arbitrator ever asks AI what it “thinks”, but whether the analysis and decision remain genuinely those of the arbitrator.
[1] UNCITRAL, Colloquium on the Use of Artificial Intelligence (AI) in Dispute Resolution and Remote Hearings in Arbitration and Mediation, 16-17 February 2026, https://uncitral.un.org/en/colloquium_aidisputeresolution (last accessed 4 September 2026).
[2] 2025 Chartered Institute of Arbitrators Guideline on the Use of AI in Arbitration, ¶ 8.1.
[3] Id. ¶ 8.2.
[4] Ibid.
[5] Id. ¶¶ 8.3-8.4.
[6] 2024 Silicon Valley Arbitration & Mediation Center Guidelines on the Use of Artificial Intelligence in Arbitration, Guideline 6.
[7] National Academy of Arbitrators, Advisory Opinion No. 28: The Arbitrator’s Use of AI, 28 March 2026, https://naarb.org/advisory-opinions/ (last accessed 4 September 2026), pt. IV.
[8] 2020 LCIA Arbitration Rules, Art. 14.8.
[9] LCIA Guidance Note for Parties and Arbitrators, ¶¶ 198-203; 2020 LCIA Arbitration Rules, Art. 14.10.
[10] 2026 ICC Arbitration Rules, Art. 44; 2026 ICC Note to Parties and Arbitral Tribunals on the Conduct of ICC Arbitration, ¶¶ 202, 209.
[11] 2026 ICC Note, ¶ 210.
[12] 2020 LCIA Arbitration Rules, Arts. 14.9-14.10, 14.14, 30.2; 2026 ICC Arbitration Rules, Art. 44(2).
[13] M.S. Abdel Wahab, Iura Novit Arbiter in International Commercial Arbitration: The Known Unknown, in N.G. Ziadé (ed.), Festschrift Ahmed Sadek El-Kosheri (2015), pp. 3-26.
[14] UNCITRAL Model Law on International Commercial Arbitration (1985, with amendments as adopted in 2006), Art. 18.
[15] Convention on the Recognition and Enforcement of Foreign Arbitral Awards, opened for signature 10 June 1958, entered into force 7 June 1959, Art. V(1)(b).
[16] SVAMC Guidelines, Guideline 7.
[17] Ciarb AI Guideline, ¶¶ 2.2-2.3.
[18] Advisory Opinion No. 28: The Arbitrator’s Use of AI, pt. III.
[19] R (Ayinde) v London Borough of Haringey; Al-Haroun v Qatar National Bank QPSC and QNB Capital LLC [2025] EWHC 1383 (Admin) [6], [15].
[20] Id. [16]; Ciarb AI Guideline, ¶¶ 2.6, 8.3-8.4; SVAMC Guidelines, Guidelines 6-7.
[21] Ciarb AI Guideline, ¶¶ 7.1, 9.1-9.2.
[22] SVAMC Guidelines, Guidelines 3 and 7.
[23] Advisory Opinion No. 28: The Arbitrator’s Use of AI, pts. II and IV.
[24] 2026 AAA AI Led Arbitration Rules, R-2(c), R-8(a).
[25] Aceris Law LLC, When Arbitrators Use AI: LaPaglia v. Valve and the Boundaries of Adjudication, 19 April 2025, https://www.acerislaw.com/when-arbitrators-use-ai-lapaglia-v-valve-and-the-boundaries-of-adjudication/ (last accessed 4 September 2026).
[26] LaPaglia v. Valve Corp., No. 3:25-cv-00833-RBM-DDL, ECF No. 18, at 12 (S.D. Cal. Dec. 9, 2025).
[27] Association des ressources intermédiaires d’hébergement du Québec (ARIHQ) c. Santé Québec – Centre intégré universitaire de santé et de services sociaux du Centre-Sud-de-l’Île-de-Montréal, 2026 QCCS 1360 at paras. 69-71, 86-87, 113, 117-122.
[28] Id. at para. 117.
[29] UNCITRAL Model Law, Arts. 34(2)(a)(ii), 34(2)(a)(iv), 34(2)(b)(ii).
[30] New York Convention, Arts. V(1)(b), V(1)(d), V(2)(b).