AI in dispute resolution: who uses it, for what, and the rules that apply

The short version

• The AI uses we found in dispute resolution are mostly support work: summaries, transcripts, research, choosing a neutral and running cases.
• Every mediation, ODR and court AI text we checked calls for human oversight of AI. Confidentiality, above all in private sessions with one party (caucuses), is a key challenge in mediation.
• Institutions run their own AI tools. AAA’s opt-in AI arbitrator drafts awards in two-party, documents-only construction cases that a human arbitrator reviews and signs.
• Under national rules applying from 20 September 2028, EU consumer ADR bodies that use automated means in decision-making must tell parties in advance and let them ask for review by a natural person.
• From 2 December 2027, the EU AI Act can treat AI that helps an ADR body decide a case with legal effects as high-risk. The European Commission’s draft guidelines of 19 May 2026, not yet adopted, leave out AI used where the outcome is a voluntary agreement or non-binding recommendation.^{1,3,5,7,8,9,10,11,12,14,16,17,19}

What is AI in dispute resolution?

AI in dispute resolution is the use of artificial intelligence (AI) tools by parties, neutrals (independent third parties such as mediators and arbitrators), the institutions and schemes that run cases, and courts’ online services. The secretariat of the United Nations Commission on International Trade Law (UNCITRAL) gives a non-exhaustive list of eight existing or potential uses, from choosing a neutral and legal research to drafting awards, predicting outcomes and case management (A/CN.9/1215, para 9).¹⁸

Alternative dispute resolution (ADR) means resolving a dispute outside the courts: in mediation a neutral helps the parties agree, and in arbitration a neutral decides. The European Commission’s draft guidelines of 19 May 2026, not yet adopted, also give conciliation, ombudsmen and complaints boards as examples (para 426). Online dispute resolution (ODR) runs a process through “electronic communications and other information and communication technology” (UNCITRAL Technical Notes, para 24). Uses range from AI-assisted, where a person decides with AI support (a human in the loop), to AI-led, where an AI proposes a decision that a human arbitrator reviews and finalises, as under AAA’s AI-led arbitration rules (R-2(c)).^10,11,17,21

This guide does not cover disputes about AI systems themselves: JAMS’s AI Disputes Clause and Rules govern those, not a neutral’s own use of AI. AAA’s Resolution Simulator (4 March 2026), which gives one party a non-binding, AI-generated simulated decision with no award, belongs in our guide to how AI is used to avoid disputes before they start.^22,52,53

How is AI used in dispute resolution today?

In the surveys and releases we found, AI in dispute resolution is used mostly for preparation and administration around a human decision. The Centre for Effective Dispute Resolution (CEDR) based its 2025 Mediation Audit on 461 responses from UK civil and commercial mediators: 64% had considered using AI and 59% had used it, most often to summarise documents and meetings. A very few used it to suggest negotiating strategies, and none in direct engagement with parties (printed pp. 1, 24, 25).¹

In arbitration, 64% of respondents to the 2025 Queen Mary University of London survey had used AI for factual and legal research in the past five years (printed pp. 27, 37). AAA’s opt-in AI arbitrator drafts awards in two-party, documents-only construction cases, and a human arbitrator reviews, may rewrite and signs each award (from 3 November 2025; R-8). Our guide sets out how AI is used in arbitration.^9,10,23

The mediation guidelines of the International Bar Association (IBA) see AI “improving efficiency, reducing costs, and broadening access to justice”, with safeguards (p. 2), and most positive views in CEDR’s audit pointed to less administrative work, such as reading time (printed p. 26). Asked what drives greater use of AI in international arbitration, survey respondents most often chose saving party and counsel time (54%) and cutting costs (44%) (Chart 19). Our legal tech guide covers the AI tools lawyers use and the rules on using them.^1,3,23

Of the 37 States (countries) that answered UNCITRAL’s questionnaire, 34 wanted work on AI in dispute resolution, particularly where AI supports people rather than deciding. Collecting evidence and case management drew the most support; AI as decision maker drew less (A/CN.9/1271, paras 12, 13).¹⁹

How is AI used in mediation?

AI in mediation helps prepare, summarise and administer cases, while the parties keep “full control over decision-making about their dispute” (IBA guidelines, Part two, 1). The IBA Mediation Committee’s guidelines (19 June 2025) list uses for each participant (the IBA guidelines in our AI rules tracker):^2,3

• Mediators: checking whether a case suits mediation, summaries and timelines, preparing questions, generating settlement options and spotting imbalance between parties (Part one, 2.1 to 2.8).
• Parties and their representatives: litigation risk analysis, choosing a mediator, position papers and first drafts of settlement agreements (3.1 to 3.6), with any legal advice from AI reviewed by a legal professional (Part two, 1.2).
• Institutions: case administration, fees, unclear dispute resolution clauses, mediator panels and attestations for enforcing settlements, including under article 4 of the Singapore Convention, the UN treaty on enforcing mediated settlements (4.1 to 4.8).³

On 28 August 2024 the Singapore International Mediation Centre (SIMC) launched an AI assistant that builds chronologies, maps who is involved, summarises disputes and can help draft an initial settlement agreement “for counsel to revise and finalise”. AAA launched AAAi Mediator Search, which uses AI to find mediators by expertise, location, language and rates (6 October 2025), and an AI Agreement to Mediate Builder (19 May 2026). At a November 2025 seminar in Kyoto reported by UNCITRAL’s secretariat, participants considered mediation “more adaptable to AI use” than arbitration, and named confidentiality, especially in caucus, a key challenge (A/CN.9/1271, para 17). Mediation bodies are also writing their own rules, compared in the [guidance table](#guidance) below.^{19,24,25,26,27,28,29}

Where does AI fit in online dispute resolution and online tribunals?

AI in online dispute resolution (ODR) sits under standards that call for human oversight, and the one public online tribunal we checked in depth keeps a person deciding. UNCITRAL’s Technical Notes (2016) define a neutral as “an individual” (para 25) and do not mention AI.^6,21,32

The ODR Standards (May 2022) of the National Center for Technology and Dispute Resolution (NCTDR) and the International Council for Online Dispute Resolution (ICODR) require human oversight of “the relative control given to human and artificial decision-making strategies” (Accountable). The two bodies’ AI guidance of 8 April 2026, for online and face-to-face work, calls for human oversight “throughout the process” (the NCTDR and ICODR guidance in our AI rules tracker).^4,5,6

British Columbia’s Civil Resolution Tribunal (CRT) is an online tribunal whose Act tells it to use “electronic communication tools” (s 2(2)(c)), defined to include programs that respond “without review by an individual at the time” (s 1). It offers a free Solution Explorer, which it does not describe as AI, and a tribunal member decides cases that do not settle.^31,32

Since 1 February 2026 the CRT’s rules bar AI-invented cases and legislation (r 8.3(8)) and evidence “created or altered by an artificial intelligence tool” (r 8.3(9)). A member can weigh such material when deciding whether a party must pay another’s legal expenses or compensation for time spent (r 10.5(4)(d), (5)(a)).³³

In England and Wales, the Online Procedure Rule Committee (OPRC) proposed in July 2025 that decisions with legal consequence should “always involve human oversight, and the option for human review” (para 168(2)). Its statement of 30 July 2026 makes a voluntary Code of Practice, with standards for “the responsible and ethical use of AI”, its immediate priority.^34,35

The Council of Europe’s European Ethical Charter on the use of AI in judicial systems (adopted 3 and 4 December 2018) says ODR users should be told whether their dispute is handled “in an entirely automated way or with the involvement of a mediator or arbitrator”. Court review “should always be considered”, especially where a user consented to fully automated ODR (Appendix II, printed p. 65).⁸

How do ombudsman and complaint schemes use AI?

The one ombudsman scheme we found describing its own AI use, the UK’s Financial Ombudsman Service (FOS), uses AI to support its service while people decide cases: “Colleagues use their own judgement and expertise to resolve cases.” Its 2026/27 plan aims to use AI “to improve speed, consistency and governance” (p. 7).^37,38

The Financial Conduct Authority (FCA), which regulates UK financial firms, does “not plan to introduce extra regulations for AI” and uses “an AI voice bot to point consumers to the correct organisation” when they first contact it (first published 8 September 2025). Firms’ complaint procedures “would include complaints about AI decisions” (AI Update, April 2024, para 3.46); our dispute avoidance guide covers how complaint-handling rules apply to AI.^39,40

We found no AI guidance from the Ombudsman Association, though its June 2026 conference programme included a session on complainants’ use of AI. Whether the EU AI Act reaches a scheme depends on whether its outcomes have [legal effects](#law).^17,41

What are the risks of AI in dispute resolution?

The main risks of AI in dispute resolution are leaks of confidential material, errors and bias, unreliable outcome predictions, over-reliance on AI, fabricated material and unequal access to tools.^3,6,18,33,36

• Confidentiality: confidential material entered into large language models “may be vulnerable to data breaches and unintended disclosure”, so anonymise it and limit it to what is needed (IBA Part two, 2.2).
• Errors and bias: outputs “may include errors, biases, cultural insensitivities, and inaccuracies” and are not a “final authority” (IBA Part two, 1.1). Tools can also invent sources, known as hallucination.
• Outcome predictions: a mediator’s reliance on AI predictions presented as likely outcomes that later prove wrong “could undermine the integrity and fairness” of mediation (A/CN.9/1215, para 59).
• Automation bias, the habit of accepting a machine’s answer uncritically: the AI advisory board of the Council of Europe’s justice efficiency commission (CEPEJ) says that in the judicial field “human control and supervision must never be lost sight of” (reflections, para 26).
• Fabricated material: in April 2026 the Superior Court of Québec annulled an arbitral award that relied on non-existent authorities (ARIHQ c. Santé Québec, 2026 QCCS 1360; how the Québec court annulled the award).
• Unequal access: no party should gain “a technological or informational advantage” (ODR Standards, Equal), and the NCTDR and ICODR checklist raises coaching a party who needs help with AI tools.^{3,5,6,18,33,36,42}

Can AI replace a mediator or decide a dispute?

No rule or guidance we checked lets AI replace a mediator. Mediation texts keep the parties in control, arbitration laws in several countries require a human arbitrator, the judiciary of England and Wales says AI must not decide a case, and EU consumer ADR law lets parties ask a person to review an automated outcome.^{3,7,12,43,44,45}

For mediators, UNCITRAL’s secretariat raises as an open question whether a mediator may rely on AI, since an agreement to mediate is an agreement “to seek a human mediator” (A/CN.9/1215, para 47). The Singapore Convention (art. 2(3)) and the UNCITRAL Model Law on Mediation (art. 1(3)) define mediation as help from “a third person or persons” who cannot impose a solution, and neither addresses AI.^18,54,55

A consistent theme in States’ answers to UNCITRAL was that “final decisions must be taken by natural persons” (A/CN.9/1271, para 14). At its 2026 session UNCITRAL kept its AI work exploratory, as the subject was “not yet sufficiently mature to warrant legislative work” (A/81/17, para 170), and asked its secretariat to keep monitoring (para 172).^19,20

For arbitrators, France (for domestic cases), Portugal and Scotland allow only a human arbitrator, and our arbitration guide asks whether AI can be one. For judges, the judiciary of England and Wales says “AI tools must not determine the outcome of a case” (Position Statement of 9 October 2026, para 10). Our guide to AI in law covers how courts and judges use AI.^7,43,44,45

What AI guidelines have mediation and ODR bodies published?

The mediation and ODR texts we found are guidance, not binding rules. They agree that a person stays responsible and differ on disclosure, as the table shows.^3,5,28

The Instituto de Certificação e Formação de Mediadores Lusófonos (ICFML) launched a practice guide series on 3 November 2025, opening with a volume on AI in mediation. Our AI rules tracker compares arbitral institutions’ texts; we found no text yet published by the International Chamber of Commerce (ICC) Commission’s task force on AI in dispute resolution.^51,56

What does the law say about AI and automated decisions in ADR?

In the EU, two texts deal specifically with automated decisions in ADR, and neither’s ADR rules apply yet: the consumer ADR directive, which covers any automated means, not only AI, and the AI Act. Directive (EU) 2025/2647 requires consumer ADR bodies that use automated means in decision-making to tell parties in advance and allow review by a natural person, under national rules applying from 20 September 2028 (art. 1 point (4)(b), art. 5). The reviewer must be a person from the ADR body who meets the expertise, independence and impartiality requirements of Directive 2013/11/EU (art. 6(1)). Decision-making covers whether to deal with a dispute and its outcome, not “purely administrative or technical tasks” (recital 21).^12,14,16

Under the EU AI Act, Annex III point 8(a) covers AI used by or for a judicial authority to research and interpret facts and the law and to apply the law to facts, “or to be used in a similar way in alternative dispute resolution”. Recital 61 limits the ADR part to outcomes with legal effects and excludes purely ancillary administrative activities. From 2 December 2027, the Act can treat such AI as high-risk (Regulation (EU) 2026/1744, art. 1 point (40)(b)).^13,14,15,16

The European Commission’s draft guidelines of 19 May 2026, not yet adopted, apply this to ADR and say the directive’s safeguards apply “in parallel with the AI Act” (para 433):¹⁷

• Both conditions must be met: the system is meant for use by an ADR body whose outcome has legal effects for the parties, and its purpose is to help that body research and interpret facts and law and apply the law to facts (para 425).
• Where the outcome is “a voluntary agreement or non-binding recommendation”, the system is outside point 8(a) (para 430).
• In consumer ADR, it is enough for legal effects that the outcome binds the trader only (para 432).¹⁷

On 23 July 2026, 16 arbitral institutions and bodies asked the Commission to keep institutions’ administrative tools outside the high-risk label (the institutions’ joint letter in our AI rules tracker). Our arbitration guide covers how the AI Act applies to arbitration, including its exceptions and deployers’ duties, and our glossary defines a provider and a deployer.⁴⁶

Outside the EU, we found few binding AI rules for ADR. California’s SB 574, approved on 30 September 2026, takes effect on 1 January 2027 and adds Code of Civil Procedure s 1282.1, under which an arbitrator “shall not delegate any part of their decisionmaking process to any generative artificial intelligence tool” ((b)(1)).^57,58

What can institutions, schemes and neutrals do about AI now?

Institutions, schemes and neutrals can act now, without waiting for final rules:^12,17

• Sort each tool by three tests: whether it affects which cases are taken on or how they end (directive, recital 21), whether it helps research and interpret facts and law and apply the law to facts (AI Act Annex III point 8(a)), and whether the outcome has legal effects (AI Act recital 61; draft guidelines para 430).
• Where automated means are used in decision-making, tell parties in advance and let them ask for review by a person, as EU consumer ADR bodies must from 20 September 2028.
• Get every participant’s written consent before AI notetaking (Resolution Institute draft), and consent before any digital tool is used (Civil Mediation Council (CMC) draft Code, para 2.8).
• Keep confidential material out of public tools, and anonymise and limit what goes in (IBA Part two, 2.2; Singapore Mediation Centre rules 2 and 3; WIPO Note, s 4).
• Use a disclosure statement, such as IBA Part three or the Singapore Mediation Centre sample (p. 4), answer participants’ questions about tools, and stop using a tool if a concern cannot be resolved (IBA Part two, 4.2).
• Put the NCTDR and ICODR pre-use questions to vendors, including data retention and deletion, breach protocols and data on bias and errors.
• Check authorities and evidence that parties produce with AI (CRT rr 8.3, 10.5).^{3,5,12,14,17,27,28,30,33,49}

What is still open on AI in dispute resolution?

The future of AI in dispute resolution depends on texts still in draft and on questions the texts we checked leave open:^17,20

• Whether a settlement reached on flawed AI advice could be refused enforcement under the Singapore Convention (art. 5(1)(e)) or the Model Law on Mediation (art. 19(1)(e)) (A/CN.9/1215, para 46).
• Whether an AI system could be the “third person” the Singapore Convention and the Model Law describe, and how far a mediator may rely on AI.
• Which scheme outcomes have legal effects under the AI Act, and what the Commission’s final guidelines will say (draft paras 430 to 432).
• What the CMC’s guidelines, Resolution Institute’s final policy, the OPRC’s Code of Practice and the ICC task force will say.
• What AI-assisted mediation achieves: we found no institution that has published data on its outcomes (how we search).^{12,17,18,20,27,29,35,51,54,55}