AI in dispute avoidance: what it does before a dispute and who decides

The short version

• Dispute avoidance means dealing with a disagreement before it becomes a legal dispute.
• AI for dispute avoidance is used to draft and review contracts, route enquiries and forecast outcomes. We found little published evidence that it prevents disputes.
• None of the dispute board rules, FIDIC principles, court pre-action rules or UK complaint-handling rules we checked mentions AI.
• Since 1 October 2024, courts in England and Wales have had an express power to order parties to engage in alternative dispute resolution (ADR).
• The American Arbitration Association (AAA) offers a Resolution Simulator (4 March 2026) that gives one party a non-binding, AI-generated simulated decision, with no human arbitrator and no award.
• The European Commission’s draft guidelines of 19 May 2026, not yet adopted, put AI used where the outcome is “a voluntary agreement or non-binding recommendation” outside the AI Act’s high-risk ADR category (para 430).^{1,4,7,10,12,15,18,19,24,25,29,30,42}

What is dispute avoidance?

Dispute avoidance, also called dispute prevention, is dealing with a problem before it turns into a legal dispute, and AI in dispute avoidance is AI used at that stage. A 2015 report for the Civil Justice Council of England and Wales said access to justice is better seen under three headings: dispute resolution, containment and avoidance (para 5.2). Containment stops disagreements that have arisen from escalating, avoidance stops legal problems arising at all, and the report proposed online evaluation to support avoidance (paras 5.3, 5.5).¹

International texts draw a similar line. The 2025 toolkit on investment disputes of the UN Commission on International Trade Law (UNCITRAL) defines dispute prevention as handling an investor’s grievance before it becomes a disagreement framed in legal terms, usually when the investor signals it will go to arbitration or court (para 2).²

A dispute board is a panel the parties to a contract usually set up when they sign it, to help them avoid or resolve disagreements (ICC Dispute Board Rules, arts 1, 3(1)). The International Chamber of Commerce (ICC) rules define a Disagreement as a difference that has “not yet become a Dispute”. It becomes a Dispute when it is formally referred to the board (art. 2(iii), (iv)).⁴

Other names include conflict avoidance in UK public construction and the Dispute Avoidance/Adjudication Board of FIDIC, the International Federation of Consulting Engineers. Our guide to how AI is used once a dispute exists covers what comes after.^6,7

How is AI used to avoid disputes?

AI is used to prevent disputes by drafting and reviewing contracts, routing enquiries and estimating outcomes. In the uses we found, a person, a human in the loop, takes the decision that follows. The benefits claimed are earlier insight and clearer clauses, and we found no source that measures how many disputes AI has helped avoid. None of the dispute board rules, FIDIC principles or court rules we read mentions AI as of 23 October 2026 (how we research), so any AI there is a tool a party chooses to use.^{4,7,10,11,18,24,25,29,30}

Can AI make a contract less likely to end in a dispute?

AI tools can draft dispute resolution clauses and flag clauses a person should review. We found little published evidence that this prevents disputes, and the drafter stays responsible. AAA’s ClauseBuilder AI takes a plain-language description of what the parties want and returns a clause built on the wording of AAA’s earlier clause tool. AAA recommends that a lawyer look at the clause before it goes into a contract.^10,12

On 30 September 2026 AAA said its library of arbitration and mediation clauses would be offered inside another company’s AI drafting tool. AAA said the aim is to help organisations “reduce ambiguity” in their clauses; it gave no evidence of fewer disputes.¹¹

One public benchmark is CUAD, a contract-review dataset presented at the NeurIPS 2021 machine learning conference. Legal experts from The Atticus Project annotated it, with over 13,000 labels across 41 clause types in more than 500 contracts. Its authors found in 2021 that the models they tested had “nascent performance”, that is, early-stage. Read how to evaluate an AI legal tool before buying it.¹²

In Ayinde, [2025] EWHC 1383 (Admin), the High Court of England and Wales held that lawyers must check AI research against authoritative sources ([7], [8]). AI that invents authorities is said to hallucinate; read what courts have done about AI-invented citations.¹³

What do dispute boards and construction contracts say about AI?

ICC’s Dispute Board Rules, FIDIC’s Golden Principles and the UK Construction Playbook do not mention AI. NEC’s own pages that we read mention it only in a notice restricting AI use of its contracts. They leave avoidance to people, usually a standing board. On our reading, AI can organise the records a board reviews, but the avoidance step itself stays with the board or the parties.^4,6,7,8

Under the ICC rules, a standing board holds regular meetings with the parties and, where site visits are relevant, at least three a year unless agreed otherwise (art. 12(1)). If it sees a potential Disagreement it “may raise this with the Parties” so they can avoid it themselves (art. 16). With all parties’ agreement it can also help informally, without being bound by those views if the matter later comes to it formally (art. 17).⁴

FIDIC’s Golden Principles require, unless the governing law conflicts, that all formal disputes go to a Dispute Avoidance/Adjudication Board (DAAB) for “a provisionally binding decision” before arbitration (GP5, p. 8). A standing board can also help parties settle issues informally before they become formal disputes (section 5.5, p. 12).⁷

NEC4 is a family of standard construction contracts. A 2025 NEC practice note sets out conflict avoidance clauses that use a panel to resolve disagreements early, and NEC4’s Option W3 provides dispute avoidance boards. NEC’s website says any use of its contracts and publications with AI needs its prior written consent and explicit endorsement.^8,9

The UK Construction Playbook (version 1.1, September 2022) tells contracting authorities, the public buyers, to adopt the appropriate provisions of the Conflict Avoidance Pledge as a standard clause in all public works contracts. Where appropriate, they should engage dispute avoidance boards “from inception to completion” (p. 50).⁶

How do complaint-handling rules apply to AI?

The UK complaint-handling rules we checked do not mention AI, but they cover complaints about it. The Financial Conduct Authority (FCA) says the complaints procedures of the firms it regulates “would include complaints about AI decisions” on providing, or failing to provide, a financial service (AI Update, April 2024, para 3.46). A consumer still dissatisfied after the firm’s answer can go to the Financial Ombudsman Service free of charge.^14,15,17,19

Firms must take reasonable steps to identify and remedy “recurring or systemic problems”, for example by finding root causes and, where reasonable, correcting them (DISP 1.3.3R). Within eight weeks the firm must send a final response or explain why it cannot yet (DISP 1.6.2R). Under the Consumer Duty, the FCA’s rules for retail customers, those customers must not face “unreasonable barriers” to making a complaint (PRIN 2A.6.2R(4)(f)), whatever the channel (PRIN 2A.6.1R(1)(b)).^15,16,17

The FCA 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).¹⁸

Social housing in England has similar rules. The Housing Ombudsman’s Complaint Handling Code, statutory since 1 April 2024, says a resident “does not have to use the word ‘complaint’” for it to be treated as one (1.3). Landlords must acknowledge a stage 1 complaint within five working days and give a full response within 10 working days of acknowledging it (6.2, 6.3).¹⁹

In the United States, firms under the supervision of the Consumer Financial Protection Bureau (CFPB) must respond to it and their other regulators in good time about consumer complaints (12 U.S.C. 5534(b)). The CFPB’s 2023 review found that each of the 10 largest US commercial banks used chatbots in customer service. It warned that a chatbot may recognise a dispute only when “specific words or syntax” trigger it and may “fail to recognize that a consumer is invoking their federal rights”.^20,21

On our reading, a chatbot that waits for the word “complaint” would sit badly with the Housing Ombudsman Code (1.3) and the Consumer Duty’s rule on barriers.^17,19

In the EU, since 2 August 2026, providers must design chatbots so that people know they are dealing with AI (art. 50(1); art. 113). Regulation (EU) 2026/1744 left that date unchanged.^22,23

What must parties do before going to court in England and Wales?

Before going to court in England and Wales, parties are expected to exchange enough information to understand each other’s positions, try to settle and consider ADR (Practice Direction on Pre-Action Conduct, para 3). Since 1 October 2024 the court can also order ADR. None of the court rules we read mentions AI. The Practice Direction says “litigation should be a last resort” (para 8).^24,25

ADR under the Practice Direction includes early neutral evaluation, an informed opinion on the dispute from a third party (para 10). Silence or refusal to take part in ADR “might be considered unreasonable” (para 11). If a party does not comply, the court can pause proceedings or impose sanctions, such as indemnity costs or extra interest against a defendant (paras 15, 16).²⁴

The Civil Procedure (Amendment No. 3) Rules 2024 (SI 2024/839), in force since 1 October 2024, added “promoting or using alternative dispute resolution” to the overriding objective, the core aim of the civil court rules (r. 3). They gave the court power to “order the parties to engage in” ADR (r. 5, new CPR 3.1(2)(o)). The court must consider ordering or encouraging ADR when it gives directions (rr. 8, 9), and failing to comply with an ADR order, or unreasonably failing to engage in ADR, counts when it decides costs (r. 10).²⁵

The rules’ Explanatory Note ties the change to Churchill v Merthyr Tydfil County Borough Council [2023] EWCA Civ 1416. The Court of Appeal held that a court can lawfully stay (pause) proceedings for, or order the parties to engage in, a non-court-based dispute resolution process ([58]). The order must not impair “the very essence” of the right to a judicial hearing and must be proportionate ([65], [74(ii)]). The process in that case was a council’s internal complaints procedure ([1]).^25,26

The Civil Justice Council’s 2023 review recommended a General Pre-Action Protocol (para 1.7). We found none in force as of 23 October 2026.^24,27

AI enters through the parties. The judiciary’s AI guidance of 31 October 2025 says chatbots “may be the only source of advice or assistance some litigants receive”. It adds that such litigants “rarely have the skills independently to verify” the output (p. 6).²⁸

Can AI forecast an outcome to help parties settle?

At least one tool gives a party a simulated outcome to inform settlement, and the parties still decide. AAA’s Resolution Simulator (4 March 2026) gives one party a non-binding, AI-generated simulated decision, with no human arbitrator and no award. AAA says parties may use it to shape their legal or negotiation strategy and that it may inform “alternative paths to resolution, such as settlement or mediation” (how AAA’s AI arbitrator works).^29,30

AAA has described its scope in two ways. The March 2026 release says it is “designed for single party use in documents-only construction disputes”. The product page, read on 23 October 2026, says it is launching for documents-only commercial and construction cases, excluding international and payer-provider (health insurer and healthcare provider) disputes. The page also says it is not an early case assessment tool.^29,30

The International Bar Association (IBA) Mediation Committee’s guidelines on generative AI in mediation (19 June 2025) list litigation risk analysis among parties’ uses of AI, such as weighing the strength of claims and the likelihood of settlement to judge whether to mediate (Part one, 3.1).³¹

Once parties settle through mediation, the Singapore Convention on Mediation provides for enforcing international commercial settlement agreements (art. 1(1)), but not consumer ones (art. 1(2)(a)). It has been in force since 12 September 2020, with 60 signatories and 25 parties according to the UN Treaty Collection as at 23 October 2026. Neither the Convention nor UNCITRAL’s 2018 Model Law on mediation addresses AI; read whether AI can act as a mediator.^32,33,34

How do states try to prevent investment disputes?

States try to prevent investment disputes through government agencies, investment ombudsmen and early warning mechanisms that catch an investor’s grievance before it becomes a claim. UNCITRAL adopted a toolkit on this by consensus on 21 July 2025 and recommended its use by states, investors and other stakeholders (A/80/17, paras 145, 146).^2,3

The final text, published in 2026, calls itself “non-prescriptive”, offering options rather than recommendations (para 4). It describes an early warning mechanism as a process to identify and flag potential issues before they escalate (para 39). It mentions live chatbots on investor portals that answer questions and direct enquiries (para 16) but does not call them AI.²

At its 2026 session UNCITRAL kept its separate 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).³⁵

What are the risks of AI in dispute avoidance?

The risks of AI in dispute avoidance named in the texts we checked are errors and bias, leaks of confidential material, reliance on AI in place of a lawyer, and disputes about AI systems themselves.^{21,31,36,38,40}

• Errors and leaks: the IBA’s mediation guidelines say legal professionals should review any legal advice from AI (Part two, 1.2). Confidential material entered into large language models “may be vulnerable to data breaches and unintended disclosure”, and anonymising it and entering only what is needed reduces that risk (Part two, 2.2).
• Bias: the Online Dispute Resolution Standards (May 2022) of the National Center for Technology and Dispute Resolution (NCTDR) and the International Council for Online Dispute Resolution (ICODR) require efforts to stop AI “creating, replicating, or compounding bias” (Equal). The two bodies’ April 2026 guidance on AI applies the standards to third parties using AI.
• Advice without a lawyer: in an order of 14 January 2025 (In the Matter of DoNotPay, Inc.), the US Federal Trade Commission barred unsupported claims that the service “operates like a human lawyer”. That includes claims that it uses legal expertise when it writes demand letters or starts small claims cases, and the order required it to pay US$193,000 (Parts I.A, III).
• Disputes about AI: JAMS’s AI Disputes Clause and Rules govern disputes about AI systems, not a neutral’s own use of AI (see AI disputes protective order). On 24 June 2026 AAA and others launched the Legal Context Protocol, an open standard rather than law, to make legal terms, consent and dispute resolution “discoverable and verifiable” when AI agents transact.^{31,36,37,38,40,41}

Does the EU AI Act cover AI used to avoid disputes?

Usually not as high-risk AI, on our reading, because most avoidance tools are run by a party on its own matters, not by a court or ADR body. From 2 December 2027, the EU AI Act can treat AI that helps a court or an ADR body decide a case with legal effects as high-risk (Annex III point 8(a); Regulation (EU) 2026/1744, art. 1 point (40)(b)). Recital 61 limits the ADR part to outcomes with legal effects and excludes purely ancillary administrative activities.^22,23

The European Commission’s draft guidelines of 19 May 2026, not yet adopted, say a system must be intended for use by an ADR body to fall under point 8(a) (para 425). Where the outcome is “a voluntary agreement or non-binding recommendation”, the system is outside point 8(a) (para 430).⁴²

Separately, providers must design chatbots so that people know they are dealing with AI (art. 50(1)), whether or not the system is high-risk. Our guide covers the AI Act and consumer ADR rules for dispute resolution bodies.²²

Open questions on AI in dispute avoidance

• Whether a dispute board decision that binds only provisionally (FIDIC GP5) has “legal effects” for point 8(a), so that AI used by a board could be high-risk (draft guidelines paras 425, 431).
• How chatbot-first complaint journeys fit the Consumer Duty’s rule against unreasonable barriers and the Housing Ombudsman Code’s rule that the word “complaint” is not needed.
• Whether AI contract-review tools may process NEC contracts, given NEC’s notice barring AI use of its texts without written permission.
• Whether the General Pre-Action Protocol that the Civil Justice Council recommended in 2023 is adopted, and in what form.^{7,8,17,19,27,42}