AI in dispute resolution: a glossary

A standard protective order attached as Appendix A to the JAMS Artificial Intelligence Disputes Rules (effective 14 June 2024). Under Rule 16.1 it applies by default to protect confidential information unless the parties agree a different order. Any production or inspection of AI systems or related materials, such as models, software and training data, is limited to experts working in a secured environment set up by the disclosing party, and the experts may not take material out of it. The order itself sets Confidential and Highly Confidential tiers and limits who may see each.^12,13

Source: JAMS AI Disputes Rules, Rule 16.1(a) and (b), and the Protective Order page (Appendix A). Paraphrased, because the page says the rules are JAMS’s copyrighted property.

See also: JAMS entry

Section 23A of the Arbitration Act 1996 was inserted by the Arbitration Act 2025 and has been in force since 1 August 2025. It requires a person approached about a possible appointment, and an arbitrator once appointed, to disclose as soon as reasonably practical any circumstances that might reasonably give rise to justifiable doubts about their impartiality. This includes circumstances they ought reasonably to be aware of. The section concerns impartiality, extends to England, Wales and Northern Ireland, and does not mention AI.^10,11

Source: Section 23A as shown on legislation.gov.uk, with its annotation (inserted 1.8.2025 by Arbitration Act 2025 c. 4, ss 2(2) and 17(2), and SI 2025/905, reg 2). The commencement date is confirmed by SI 2025/905, reg 2, which brings the 2025 Act into force on 1 August 2025.

See also: AI rules tracker, disclosure column

A person, company, public authority or other body that uses an AI system under its own authority, unless the use is part of a personal non-professional activity.^6,7

Source: Regulation (EU) 2024/1689, Article 3, point (4). Not amended by Regulation (EU) 2026/1744.

See also: Provider

Output that an AI tool makes up and presents as if it were true, such as a case, quotation, statute or fact that does not exist. The Ciarb guideline defines the term along these lines and urges caution with hallucinated material such as academic sources, legislation or evidence.^3,4,5

Source: Ciarb Guideline on the Use of AI in Arbitration (2025, updated September 2025), defined terms (PDF p. 4) and para 2.6. Paraphrased, because Ciarb’s text is ‘All rights reserved’. For descriptions of the risk see 2026 QCCS 1360 at [90] and [2025] EWHC 1383 (Admin) at [6]. For examples in practice see 2026 QCCS 1360 at [103] to [113] and [2025] EWHC 1383 (Admin) at [36] to [38] and [74].

See also: Ciarb Guideline, Awards challenged over AI use

Under Article 6(2) and Annex III point 8(a) of the EU AI Act, an AI system is high-risk if it is intended to be used by a judicial authority, or on its behalf, to assist it in researching and interpreting facts and the law and in applying the law to a concrete set of facts, or to be used in a similar way in alternative dispute resolution. Article 6(3) lets such a system fall outside the label where it does not pose a significant risk of harm, including by not materially influencing the outcome of decision-making, and one of four listed conditions is met. Regulation (EU) 2026/1744 moved the date from which Chapter III, Sections 1 to 3 (classification, requirements and obligations) apply to Annex III systems from 2 August 2026 to 2 December 2027.^6,7,8,9

Source: Regulation (EU) 2024/1689, Article 6(2), Article 6(3), Annex III point 8(a) and Article 113. Regulation (EU) 2026/1744 of 8 July 2026 (published 24 July 2026), Article 1, point (40), which replaces point (c) of the third paragraph of Article 113. The Omnibus did not change the wording of Annex III point 8(a).

See also: Narrow procedural task, Preparatory task, LCIA entry

A way of working in which a person reviews what an AI system produces, can override it, and takes responsibility before it affects a decision. None of the instruments we checked defines the phrase. The closest texts are the EU AI Act’s human oversight rule, which includes the power to disregard or override output (Article 14(4)), Recital 61, which says final decision-making in justice must remain a ‘human-driven activity’, and Ciarb’s advice that arbitrators verify AI output and take responsibility for the award (paras 8.3 to 8.4).^4,6,7

Source: Our own definition. It draws on EU AI Act Article 14(4) and Recital 61, and on Ciarb Guideline paras 8.3 to 8.4 (paraphrased).

See also: Courts and Tribunals Judiciary position statement

In the EU AI Act, the test Recital 61 uses to explain when AI used by alternative dispute resolution bodies should count as high-risk: when the outcome of the ADR proceeding produces legal effects for the parties. The Act does not define the phrase in Article 3. The binding text in Annex III point 8(a) instead covers AI ‘used in a similar way in alternative dispute resolution’.^6,7

Source: Recital 61 and Annex III point 8(a) of Regulation (EU) 2024/1689. The phrase also appears, in a different context, in Article 86 (right to an explanation of individual decisions).

See also: High-risk AI system in the administration of justice

One of four conditions in Article 6(3) of the EU AI Act (point (a)) under which an AI system listed in Annex III is not treated as high-risk, provided it does not pose a significant risk of harm, including by not materially influencing the outcome of decision-making. Recital 53 gives examples such as turning unstructured data into structured data, sorting incoming documents into categories, and spotting duplicates among many applications. The exemption never applies where the system profiles natural persons, and a provider relying on it must document its assessment before putting the system on the market (Article 6(3) and (4)).^6,7

Source: Regulation (EU) 2024/1689, Article 6(3)(a), Article 6(4) and Recital 53. The Omnibus inserted Article 6(1a) to (1c) but did not amend Article 6(3).

See also: Preparatory task

The principle that an arbitrator’s decision-making mandate is personal and cannot be handed to anyone else, including an AI tool. AI may help with tasks, but the arbitrator must carry out their own analysis of the facts, the law and the evidence and remains responsible for any errors in the result.^1,2,3

Source: SVAMC Guideline 6 (p. 12) and its commentary (pp. 19 to 20). VIAC Note paras 2.1 and 2.2. The Superior Court of Québec applied the same principle when it annulled an award in 2026 QCCS 1360 at [86] and [114] to [116].

See also: SVAMC Guidelines, VIAC Note, Awards challenged over AI use

Under Article 6(3)(d) of the EU AI Act, an AI system listed in Annex III is not treated as high-risk where it only performs a task preparatory to an assessment relevant to an Annex III use, subject to the same conditions as the other Article 6(3) exemptions. Recital 53 gives examples of file handling, such as indexing, searching, text and speech processing, linking data to other sources, and translating initial documents.^6,7

Source: Regulation (EU) 2024/1689, Article 6(3)(d) and Recital 53.

See also: Narrow procedural task

The first procedural order in an arbitration, which usually records the timetable and the procedural ground rules agreed at the start of the case. Both VIAC and Ciarb point to it as a place to record what the parties and tribunal have agreed about AI use. VIAC suggests agreeing provisions on confidentiality and transparent AI use there, and Ciarb’s template AI order can be built into PO 1 or terms of reference.^2,4

Source: VIAC Note para 5.3 (which mentions agreeing AI provisions ‘e.g. in Procedural Order No. 1’). Ciarb Guideline para 4.6 (record decisions on AI in a procedural order) and the drafting note to the Appendix B template order (PDF p. 25), paraphrased.

See also: VIAC Note, Ciarb Guideline

A person, company, public authority or other body that develops an AI system or general-purpose AI model, or has one developed, and places it on the market or puts it into service under its own name or trademark, whether paid or free.^6,7

Source: Regulation (EU) 2024/1689, Article 3, point (3). Not amended by Regulation (EU) 2026/1744.

See also: Deployer

Under Recital 61 of the EU AI Act, the high-risk label for justice should not extend to AI used only for back-office work that does not affect how justice is administered in individual cases. The recital’s examples are anonymising or pseudonymising decisions, documents or data, communication between staff, and administrative tasks.^6,7

Source: Recital 61 of Regulation (EU) 2024/1689. The recital sits in the preamble, not in the numbered articles. The list of high-risk uses itself is in Annex III, read with Article 6.

See also: Legal effects

An application to a court at the seat of arbitration to cancel an award. Under Article 34 of the UNCITRAL Model Law it is the only recourse against an award. It is limited to listed grounds, such as a tribunal composition or arbitral procedure that did not follow the parties’ agreement (Article 34(2)(a)(iv)), and must normally be brought within three months of receiving the award (Article 34(3)). Québec’s Code of Civil Procedure, which calls it annulment, follows the same pattern in articles 646 and 648, and the court applied those articles in 2026 QCCS 1360.^3,14,15

Source: UNCITRAL Model Law (1985, with 2006 amendments), Article 34. Code of Civil Procedure (Québec), arts 646 and 648, applied in 2026 QCCS 1360 at [33] to [38], [76] and [116].

See also: Vacatur, Awards challenged over AI use

A US court order cancelling an arbitration award. Section 10(a) of the Federal Arbitration Act lists four grounds on which a court may vacate an award: an award procured by corruption, fraud or undue means, evident partiality or corruption in the arbitrators, misconduct such as refusing to hear pertinent and material evidence, or arbitrators exceeding or imperfectly executing their powers. A federal court also needs its own basis for jurisdiction to hear a section 10 petition, which is why the LaPaglia petition was dismissed without any ruling on its AI ground.^16,17

Source: 9 U.S.C. section 10(a) (U.S. Code 2024 edition on govinfo, laws in effect as of 6 January 2025). The jurisdiction point comes from the LaPaglia order (Doc. 18, pp. 7 to 12), applying Badgerow v. Walters.

See also: Set aside, Awards challenged over AI use