AI in Arbitration: Frameworks, Applications, and Challenges

Legal and Regulatory Frameworks

The incorporation of AI into arbitration raises questions about how existing laws and regulations apply to its use. Globally, no uniform or comprehensive legal regime yet governs AI in arbitration, but several jurisdictions have started to address the intersection of AI and dispute resolution through legislation, regulations, or policy guidelines.

In February 2026, UNCITRAL Working Group II held a colloquium on the use of AI in dispute resolution and on remote hearings in arbitration and mediation. The resulting report (A/CN.9/1237) identifies issues for possible future procedural guidance, but UNCITRAL has not adopted a uniform treaty, model law, or binding rule governing AI use in arbitration.

Convention on the Recognition and Enforcement of Foreign Arbitral Awards (New York Convention)

A. Overview of the Convention’s Scope

Article I(1) states that the Convention applies to the “recognition and enforcement of arbitral awards made in the territory of a State other than the State where the recognition and enforcement of such awards are sought,” arising out of disputes between “persons, whether physical or legal.”

The Convention does not define “arbitrator.” It refers to awards made by arbitrators appointed for each case and to awards made by permanent arbitral bodies. It does not address whether a non-human system could itself be the legally responsible adjudicator.

B. Key Provisions Envision Human Arbitrators

Article II concerns the written arbitration agreement. The Convention’s assumptions about the adjudicator appear most clearly in Article V, which addresses notice of the appointment of the arbitrator, the composition of the arbitral authority, and compliance with the parties’ agreed procedure.

Articles III–V: Concern the recognition/enforcement of awards and set out grounds upon which enforcement may be refused. For instance, Article V(1)(b) refers to a party “not [being] given proper notice of the appointment of the arbitrator,” or “otherwise unable to present his case.”

Article V(1)(d): Allows refusal of enforcement if “the composition of the arbitral authority … was not in accordance with the agreement of the parties….” The provision presupposes an identifiable arbitral authority constituted under the parties’ agreement and the applicable law, although it does not expressly define every permissible form of decision-maker.

Article I(2): The “term ‘arbitral awards’ shall include not only awards made by arbitrators appointed for each case but also those made by permanent arbitral bodies….” This language recognizes institutional or permanent forms of arbitration; it does not itself authorize software to act as the adjudicator. The legal status of the tribunal remains governed by the applicable arbitration law and the parties’ agreement.

C. The Convention’s Enforcement Regime Presupposes Human Judgment

The enforcement structure requires, among other things, a valid arbitration agreement, proper notice, an opportunity to present the case, and a tribunal constituted in accordance with the parties’ agreement or the applicable law. Article V(1)(a) addresses party capacity and validity of the agreement; Article V(1)(b) addresses notice and the ability to present the case; and Article V(1)(d) addresses composition and procedure. A fully autonomous system would create unresolved questions about valid constitution, challenge procedures, legal responsibility, and attribution of the award.

D. “Permanent Arbitral Bodies” Do Not Imply Autonomous AI

While Article I(2) recognizes awards made by “permanent arbitral bodies,” it does not recognize an autonomous AI system as an arbitrator. Any attempt to use AI as the sole decision-maker would still have to satisfy the law of the seat, the parties’ agreement, due-process requirements, and the enforcement standards of the Convention.

UNCITRAL Model Law on International Commercial Arbitration

A. Terminology and Structure

Article 2(b) of the Model Law defines “arbitral tribunal” as “a sole arbitrator or a panel of arbitrators.” Articles 10 and 11 address the number and appointment of arbitrators, and Article 12(1) begins: “When a person is approached in connection with his possible appointment as an arbitrator….” The structure therefore presupposes an appointed person or panel exercising the arbitral office.

B. Core Provisions That Imply a Human Arbitrator

Articles 11–15 on appointment, disclosure, challenge, termination, and replacement place personal duties on the appointed arbitrator. Article 12(1) requires disclosure of circumstances likely to give rise to justifiable doubts as to impartiality or independence. A provider may disclose a system’s design, data practices, or limitations, but that is not the same as the appointed arbitrator satisfying the Model Law’s personal disclosure duty.

Article 31(1) requires the award to be in writing and signed by the arbitrator or arbitrators. Electronic signatures may satisfy applicable form requirements, but the signature must remain legally attributable to the appointed arbitrator or arbitrators, who are responsible for the award. An autonomous system cannot, by itself, assume that statutory office or accountability.

Article 19 affirms the freedom of the parties to determine procedure, but absent party agreement, the tribunal “may conduct the arbitration in such manner as it considers appropriate.” This includes evaluating evidence, hearing witnesses, and ensuring fundamental fairness (Articles 18, 24). That discretionary, human-like judgment is not accounted for if the “tribunal” were simply an AI tool with no human oversight.

C. Arbitrator’s Duties Presuppose Personal Judgment

Many of the Model Law’s articles require the arbitrator to exercise personal discretion and to do so impartially:

• Article 18: “The parties shall be treated with equality and each party shall be given a full opportunity of presenting his case.” Human arbitrators are responsible for ensuring this fundamental right.
• Article 24: The tribunal hears the parties, manages documents, questions witnesses, etc.
• Article 26: The tribunal may appoint experts and question them.
• Article 17 (and especially the 2006 amendments in Chapter IV A) require the arbitrator to assess whether an “interim measure” is warranted, including whether “[h]arm not adequately reparable by an award of damages is likely to result if the measure is not ordered” (art. 17 A(1)(a)).

These duties reflect a legal expectation of personal capacity for judgment, integral to the role of “arbitrator” as recognized by the Model Law.

United States

The United States has no specific federal statute or set of arbitration rules explicitly regulating the use of AI in arbitral proceedings. The Federal Arbitration Act (“FAA”), first enacted in 1925 (now codified at 9 U.S.C. §§ 1–16 and supplemented by Chapters 2 and 3), was drafted with human decision-makers in mind. Indeed, its provisions refer to “the arbitrators” or “either of them” in personal terms. Fully autonomous AI “arbitrators” were obviously not contemplated in 1925.

The FAA imposes no direct ban on an arbitrator’s use of technology. Under U.S. law, arbitration is fundamentally a matter of contract, so the parties’ agreement may authorize broad use of AI-assisted tools. Consent to the use of tools, however, is not the same as legal recognition of software as the arbitrator. Whether a fully autonomous AI could issue an enforceable award remains unresolved and would depend on the agreement, the law of the seat, and the statutory grounds for vacatur and enforcement.

Where the FAA does not govern or does not preempt, state arbitration law may impose additional requirements. Any proposal for an autonomous AI decision-maker must therefore also be tested against the law of the seat, the parties’ appointment mechanism, and the forum’s definition of a valid arbitral award.

AI and the “Essence” of Arbitration Under the FAA

The threshold issue is not whether an arbitrator may use AI, but whether its use remains consistent with the parties’ arbitration agreement and with the arbitrator’s obligation to decide the dispute. Under 9 U.S.C. § 10(a), an award may be vacated on specified grounds, including corruption, fraud or undue means, evident partiality, procedural misconduct, and excess of powers. If AI supplants rather than assists the appointed arbitrator, a challenger may invoke one or more of those grounds, but AI involvement does not itself establish vacatur.

A. Replacing the Arbitrator Entirely

If the parties selected a human neutral, replacing that neutral with an autonomous system could be characterized as a departure from the agreed method of arbitration or as an excess of authority. The FAA assumes that awards are made by arbitrators, but no reported federal merits decision has recognized an autonomous AI system as the legally appointed arbitrator.

The relevant formal questions include whether the tribunal was properly appointed, whether the parties received the process for which they contracted, and whether a written, final, and legally attributable award was issued. The FAA does not impose a general requirement that an arbitrator be sworn before hearing testimony.

If an AI system controls the proceeding and the appointed arbitrator exercises no genuine judgment, a party could argue that the arbitrator exceeded the powers conferred by the agreement under 9 U.S.C. § 10(a)(4). The separate phrase “so imperfectly executed them that a mutual, final, and definite award … was not made” addresses defects in finality or definiteness, not every instance of improper delegation.

B. Public Policy Concerns

For domestic awards governed by the FAA, public policy is not a freestanding ground for vacatur: Hall Street Associates, L.L.C. v. Mattel, Inc., 552 U.S. 576, 584 (2008), treats the statutory grounds in §§ 10 and 11 as exclusive. For foreign or non-domestic awards governed by the New York Convention, Article V(2)(b) separately permits refusal of recognition or enforcement when enforcement would be contrary to the forum state’s public policy.

Potential Legal Challenges When AI Is Used in Arbitration

A. Due Process and Fair Hearing

Right to Present One’s Case (9 U.S.C. § 10(a)(3)): Both parties must have a fair opportunity to present evidence and argument. If an AI process causes the tribunal to refuse pertinent and material evidence or otherwise prejudices a party’s rights, the resulting conduct may be tested under the statute’s procedural-misconduct ground.

Transparency: Arbitrators are not generally required to reveal internal deliberations. Disclosure may nevertheless become relevant if AI introduces material facts, authorities, calculations, or theories outside the record, or if a procedural order, institutional rule, professional obligation, or the parties’ agreement requires it. The central due-process question is whether the parties had a fair opportunity to address material grounds on which the tribunal relied.

“Undue Means” (9 U.S.C. § 10(a)(1)): The statute requires that the award have been “procured by corruption, fraud, or undue means.” Mere undisclosed use of AI, without deception or misconduct connected to procuring the award, is unlikely by itself to satisfy that ground.

B. Algorithmic Bias and Fairness of Outcomes

Bias in AI Decision-Making: AI tools can reproduce or amplify bias in training data, design, or deployment. A biased damages model or evidence-ranking tool may distort the tribunal’s analysis, but the legal consequence depends on how the tool was used, whether the arbitrator independently reviewed the output, and whether a statutory ground for vacatur is established.

Challenge via “Evident Partiality” (9 U.S.C. § 10(a)(2)): Bias in a tool is not automatically “evident partiality” of the arbitrator. A successful challenge would need evidence connecting the problem to the arbitrator’s partiality, undisclosed conflict, refusal to exercise independent judgment, or another recognized statutory ground.

FAA Vacatur or Modification of AI-Assisted Awards

A. Exceeding Powers or Improper Delegation (9 U.S.C. § 10(a)(4))

An award may be vulnerable if the arbitrator effectively delegates the decision to AI and merely rubber-stamps its output, particularly where the agreement calls for personal adjudication by the selected neutral. The challenge must still be tied to the powers conferred by the agreement and satisfy the demanding standard of 9 U.S.C. § 10(a)(4).

B. Procedural Misconduct and Prejudice (9 U.S.C. § 10(a)(3))

AI use may support a procedural-misconduct argument if it introduces material information outside the record, causes the tribunal to refuse pertinent evidence, or otherwise prejudices a party’s opportunity to present its case. Under 9 U.S.C. § 10(a)(3), the challenger must show recognized misconduct or misbehavior and material prejudice, not merely that the tool made an error.

C. Narrow Scope of Review

Judicial review under the FAA is strictly limited. Under Hall Street, the statutory grounds in 9 U.S.C. §§ 10 and 11 are exclusive for proceedings governed by those provisions. Simple factual or legal errors—even if AI-related—ordinarily do not suffice. A challenger must connect the AI use to a recognized ground such as prejudicial refusal to hear pertinent evidence, evident partiality of the arbitrator, fraud in procuring the award, or excess of powers.

Recent litigation illustrates the issue but has not resolved it. In LaPaglia v. Valve Corp., No. 3:25-cv-00833-RBM-DDL (S.D. Cal. Dec. 9, 2025), the petitioner alleged that an arbitrator had “outsourc[ed] his adjudicative role to artificial intelligence.” The court dismissed for lack of subject-matter jurisdiction and did not reach the merits. The order therefore does not establish when AI-assisted analysis or drafting constitutes excess of powers or procedural misconduct.

D. Modification of Awards (9 U.S.C. § 11)

If AI introduced a clear numerical error or a clerical-type mistake in the award, courts may modify or correct rather than vacate. Such errors include “evident material miscalculation” (§ 11(a)) or defects in form not affecting the merits (§ 11(c)). These are minor and straightforward fixes.

Federal AI Policy and the 2022 AI Bill of Rights

The White House’s Blueprint for an AI Bill of Rights (October 2022) is a historical, non-binding policy framework for automated systems. Two of its principles remain relevant by analogy: “Notice and Explanation,” which favors notice that an automated system is being used and an understandable account of its role, and “Human Alternatives, Consideration, and Fallback,” which favors access to human review where appropriate.

Federal executive policy changed materially in 2025. Executive Order 14179 of January 23, 2025 directed a policy focused on American AI leadership, and OMB Memoranda M-25-21 and M-25-22 of April 3, 2025 govern federal-agency use and procurement of AI. None of those instruments directly regulates private arbitration. The 2022 Blueprint can still inform voluntary safeguards, but it is not controlling law. Current arbitration guidance—including the published 2024 SVAMC Guidelines—likewise emphasizes competence, confidentiality, independent verification, context-specific disclosure, and non-delegation of the arbitrator’s decision-making responsibility.

European Union AI Regulation

The AI Act (Regulation (EU) 2024/1689), as amended through Regulation (EU) 2026/1744, lays down harmonized rules for the development, placing on the market, putting into service, and use of AI systems in the Union. Most of the Act began applying on August 2, 2026. For high-risk systems listed in Annex III, however, the requirements in Chapter III, Sections 2 and 3 are scheduled to apply from December 2, 2027.

The relevant operative provision is Article 6(2) read with Annex III, point 8(a). It covers AI systems intended to be used by or on behalf of a judicial authority to assist in researching and interpreting facts and law and applying law to concrete facts, or to be used in a similar way in alternative dispute resolution. Recital (61) explains that this covers alternative dispute resolution proceedings whose outcomes produce legal effects for the parties and excludes purely ancillary administrative activities.

Accordingly, an AI system is not high-risk merely because it is used somewhere in an arbitration. Classification turns principally on the provider’s intended purpose and the function performed. A system intended to assist a tribunal in applying facts and law to a binding outcome can fall within Annex III, point 8(a); an administrative tool that does not materially influence adjudication ordinarily will not.

I. Conditions Triggering High-Risk Classification for AI Arbitration

Does the AI “assist” in interpreting facts or law?

Under Article 6(2) and Annex III, point 8(a), the relevant question is whether the system is intended to assist in researching and interpreting facts and law and applying law to a concrete set of facts in a judicial or similar ADR function.

A tool intended to provide reasoned recommendations on liability, damages, or dispositive legal and factual findings is likely within that function. Predictive analytics or procedural recommendations require a closer assessment: the decisive issue is whether the intended use materially influences adjudicative decision-making rather than merely supporting an ancillary task.

Does the AI system produce legal effects?

Annex III, point 8(a) applies to comparable use in alternative dispute resolution. Recital (61) confirms that this covers proceedings whose outcomes produce legal effects for the parties.

Arbitral awards are ordinarily binding and may be enforced by national courts, so their outcomes produce legal effects. Even so, the system is captured only when its intended purpose is to assist the legal or factual application described in Annex III, point 8(a).

Exclusion of “purely ancillary” uses

Recital (61) excludes “purely ancillary administrative activities that do not affect the actual administration of justice in individual cases.” Article 6(3) also provides that an Annex III system is not high-risk where it does not pose a significant risk of harm and does not materially influence the outcome of decision-making, including where it performs a narrow procedural, preparatory, pattern-detection, or prior-result-improvement task. Systems that profile natural persons remain high-risk.

Scheduling, formatting, transcription, routing, and anonymization will generally fall outside the high-risk category when they are genuinely ancillary. The dividing line is the system’s intended purpose and whether its output materially influences how facts and law are applied to the dispute.

II. Legal and Practical Implications for AI in Arbitration in EU

When an arbitration-related system is classified as high-risk under Annex III, the Act’s provider and deployer obligations will apply on the amended timetable—currently December 2, 2027 for the relevant Chapter III requirements. Those requirements include risk management, data governance, technical documentation, logging, transparency, human oversight, accuracy, robustness, cybersecurity, conformity assessment, and post-market monitoring.

A. Risk Management System (Article 9)

Providers of high-risk AI systems are required to implement a documented, continuous risk management process covering the system’s entire lifecycle. For AI arbitration, the provider of the software (i.e. the entity placing it on the market or putting it into service) must:

• Identify potential risks (including the risk of incorrect, biased, or otherwise harmful award recommendations).
• Mitigate or prevent those risks through corresponding technical or organisational measures.
• Account for reasonably foreseeable misuse (for instance, using the tool for types of disputes or jurisdictions it is not designed to handle).

B. Data Governance and Quality (Article 10)

Data sets used to train, validate, or test a high-risk AI system must:

• Be relevant, representative, and correct to the greatest extent possible.
• Undergo appropriate governance and management to reduce errors or potential biases that could lead to discriminatory decisions or outcomes in arbitration.

C. Technical Documentation, Record-Keeping, and Logging (Articles 11 and 12)

High-risk AI systems must include:

• Clear, up-to-date technical documentation, covering the model design, data sets, performance metrics, known limitations, and other key technical aspects.
• Proper record-keeping (“logging”) of the system’s operations and outcomes, enabling traceability and ex post review (e.g. in the event of challenges to an arbitral decision relying on the AI’s outputs).

D. Transparency and Instructions for Use (Article 13)

Providers of high-risk AI systems must ensure sufficient transparency by:

• Supplying deployers—including, depending on the use case, arbitral institutions, tribunals, law firms, or parties—with instructions that identify the system’s intended purpose, capabilities, limitations, expected accuracy, foreseeable risks, required input specifications, and human-oversight measures.
• Providing information that enables the deployer to interpret and use the output appropriately, including known circumstances that may affect performance and the technical and organisational measures needed for safe operation. Article 13 does not require a universal disclosure of model internals or confidence scores in every case.

E. Human Oversight (Article 14)

High-risk AI systems must be designed and developed to allow for human oversight:

• The system must be designed so that the natural persons assigned to oversight can understand its capacities and limitations, monitor its operation, detect anomalies, disregard or override outputs, and interrupt or stop the system where necessary.
• In arbitration, the practical consequence is that the appointed arbitrator or tribunal should retain genuine control over procedure, evidence, legal analysis, and the award. Human oversight under the AI Act complements—but does not replace—the independent duties imposed by arbitration law.

F. Accuracy, Robustness, and Cybersecurity (Article 15)

Providers must ensure that high-risk AI systems:

• Achieve and maintain a level of accuracy that is appropriate in relation to the system’s intended purpose (e.g. suggesting case outcomes in arbitration).
• Are sufficiently robust and resilient against errors, manipulation, or cybersecurity threats—particularly critical for AI tools that could otherwise be hacked to produce fraudulent or manipulated arbitral results.

G. Post-Market Monitoring (Article 72)

Providers of high-risk AI systems must establish and document a post-market monitoring system proportionate to the technology and risks.

• That system must actively and systematically collect and analyse relevant performance data throughout the system’s lifetime, including information from deployers and other sources where appropriate.
• Providers must use the results to maintain compliance and take corrective action when performance, safety, bias, or other compliance problems emerge.

III. The Role of the Provider vs. the Arbitration Institution (Deployer)

Pursuant to Article 3(3) of Regulation (EU) 2024/1689, a provider is defined as:

“‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge.”

Accordingly, a law firm, software vendor, arbitral institution, or specialist developer can be a provider if it develops an AI system—or has one developed—and places it on the market or puts it into service under its own name or trademark. Providers of high-risk systems must comply with Article 16 and the related provisions, ensure conformity with Articles 9–15, complete the applicable conformity assessment under Article 43, register the system where required, and maintain post-market monitoring under Article 72.

Under Article 3(4) of the AI Act, a deployer is defined as:

“‘deployer’ means a natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity.”

In arbitration, the deployer is the person or body using the system under its authority. Depending on the workflow, that may be an arbitral institution, tribunal, law firm, party, or other participant. Article 26 requires deployers of high-risk systems, among other things, to follow the instructions for use, assign competent human oversight, monitor operation, act on relevant risks, and retain automatically generated logs within their control for the prescribed period.

IV. Distinguishing “High-Risk” vs. “Ancillary” AI in Arbitration

Article 6(2) and Annex III, point 8(a) identify the adjudicative use case; Article 6(3) then provides a limited exclusion where the system does not pose a significant risk of harm and does not materially influence the outcome of decision-making. Recital (61) illustrates that purely ancillary administrative activities are outside the adjudicative high-risk category.

Therefore:

High-risk arbitration AI:

• Covered by Article 6(2) and Annex III, point 8(a), subject to Article 6(3), where the system is intended to assist an arbitral tribunal in researching or interpreting facts and law or in applying law to concrete facts in a proceeding whose outcome has legal effects.
• Examples may include systems intended to recommend dispositive legal conclusions, evaluate core evidence, advise on liability or quantum, or draft substantive portions of the award for tribunal adoption. Classification remains purpose- and function-specific.

Not high-risk:

• A tool used only for a narrow procedural, preparatory, or administrative task—such as scheduling, document formatting, transcription, routing, or anonymization—where it does not materially influence the legal or factual determination.
• Such a system may fall outside Annex III, point 8(a), or within the Article 6(3) exclusion, depending on its intended purpose and actual decision-making role. Profiling of natural persons is treated differently and remains high-risk.

Boundary scenarios:

• A tool described as “supporting” may still be high-risk if its intended function materially steers the tribunal’s assessment of evidence, law, liability, damages, or other dispositive issues. Conversely, incidental use of a general-purpose tool does not automatically determine the legal classification of the underlying AI system.

The practical distinction is therefore not simply “AI used” versus “AI not used.” The classification requires a documented assessment of intended purpose, the system’s role in the proceeding, the legal effects of the outcome, the Article 6(3) criteria, and any profiling function.

The AI Act does not displace other applicable law. In particular, the GDPR may govern personal data used in prompts, training, retrieval, logging, or outputs; professional-secrecy and confidentiality duties may be stricter; and cross-border transfers, cybersecurity, intellectual property, and contractual restrictions must be assessed separately.

National Arbitration Laws in Key EU Member States

Arbitration in the EU is primarily governed by the national laws of each Member State. Over the past decades, many European countries have modernized their arbitration statutes, often drawing inspiration from the UNCITRAL Model Law on International Commercial Arbitration. However, the extent of Model Law adoption and procedural frameworks varies by country.

Summary

Germany. Arbitration is governed by Book 10 of the Code of Civil Procedure (ZPO), §§ 1025–1066. Book 10 was enacted on the basis of the 1985 UNCITRAL Model Law, with modifications. Domestic awards require a declaration of enforceability under § 1060, while foreign awards are recognized and enforced under § 1061 and the New York Convention, subject to limited refusal grounds.

France. Arbitration is governed by Articles 1442–1527 of the Code of Civil Procedure. France has not adopted the Model Law and maintains an independent domestic and international arbitration regime. Awards require exequatur for enforcement; annulment and refusal grounds are limited, and international arbitration is governed by Articles 1504–1527.

Netherlands. Arbitration is governed by Book 4 (the Arbitration Act) of the Code of Civil Procedure. There is no formal adoption of the Model Law, although the regime is materially influenced by it. Domestic awards require leave for enforcement under Article 1062, and foreign awards are addressed by Articles 1075–1076 and applicable treaties, including the New York Convention.

Sweden. Arbitration is governed by the Swedish Arbitration Act (1999:116), as amended. The Act is not a formal Model Law adoption but is substantively similar in important respects. Swedish awards are subject to limited invalidity and set-aside grounds, and foreign awards are recognized under §§ 53–60 and the New York Convention.

Spain. Arbitration is governed by Ley 60/2003, de Arbitraje, which is substantially based on the UNCITRAL Model Law. Awards are final and enforceable, with annulment available only on limited statutory grounds; foreign awards are recognized under the New York Convention and Spanish procedural law.

Italy. Arbitration is governed by Articles 806–840 of the Code of Civil Procedure. Italy has not formally adopted the Model Law, although the regime shares several Model Law principles. Domestic awards are declared enforceable by the court (tribunale) at the seat under Article 825, and foreign awards are recognized and enforced by the Court of Appeal under Articles 839–840 and the New York Convention.

AI Arbitration in Germany

Below is a high-level analysis of how AI-based arbitration fits within the wording of the German Arbitration Act in Book 10 of the Code of Civil Procedure (ZPO). The Act does not specifically regulate AI, so the analysis distinguishes permitted use of AI tools from the much more difficult question of an autonomous AI serving as arbitrator.

A. No Direct Prohibition on AI

Book 10 contains no express prohibition on an arbitrator, party, or institution using AI as a tool. Its provisions on the arbitral office, however, repeatedly presuppose an appointed person who can disclose circumstances, be challenged or replaced, conduct the proceedings, and sign the award. The absence of an AI-specific prohibition should not be treated as affirmative authorization for a non-human sole arbitrator.

B. Potential Impediments to AI as Sole Arbitrator

Sections 1036–1038 govern disclosure, challenge, and termination where an arbitrator is unable to perform the office or fails to act. Those duties attach to the person approached or appointed as arbitrator and depend on personal independence, impartiality, capacity, and accountability. An AI vendor can disclose system risks, but the software cannot itself discharge the statutory office in the same way.

Technical transparency, testing, and bias controls can make AI assistance safer, but they do not replace the appointed arbitrator’s duties under §§ 1036–1038. A human arbitrator who uses AI remains responsible for conflicts, procedural fairness, evidence assessment, legal reasoning, and the final decision.

C. Formal Requirements Under § 1035

Section 1035 gives the parties substantial freedom to agree on the appointment procedure and provides a court mechanism where appointment fails. That autonomy does not eliminate the statutory requirements attached to the arbitral office. It would therefore be unsafe to assume that naming an AI platform is valid merely because no party objects; the legal validity and enforceability of such an appointment remain untested and highly doubtful.

D. Procedural Rights: “Notice,” “Hearing,” and “Legal Audi Alteram Partem”

Section 1042 requires equal treatment and a full opportunity for each party to present its case. Section 1054 requires a written, signed award and, unless the parties agree otherwise or the award records an agreed settlement, reasons. AI assistance is compatible with those provisions only if the tribunal preserves the agreed process, considers the record, gives the parties a fair opportunity to address material issues, and independently adopts the reasoning and award.

E. Setting Aside and Public Policy

Section 1059 permits setting aside where a party lacked proper notice or was otherwise unable to present its case (§ 1059(2)(1)(b)), where the composition of the tribunal or procedure was not in accordance with the agreement or mandatory law (§ 1059(2)(1)(d)), or where recognition or enforcement would conflict with German public policy (§ 1059(2)(2)(b)). An AI-assisted award is not invalid merely because AI was used, but opaque substitution of the tribunal, material outside-the-record analysis, or denial of the right to be heard could engage those grounds.

France

A. Domestic Arbitration: Article 1450 Requires a Human Arbitrator

Article 1450 (Titre I, Chapitre II) provides, in part:

“La mission d’arbitre ne peut être exercée que par une personne physique jouissant du plein exercice de ses droits. Si la convention d’arbitrage désigne une personne morale, celle-ci ne dispose que du pouvoir d’organiser l’arbitrage.”

This is the single most direct statement in the Act that speaks to who (or what) may serve as arbitrator. It clearly states that the function of deciding the case (“la mission d’arbitre”) can only be carried out by a “personne physique” (a natural person) enjoying full civil rights. Meanwhile, a personne morale (legal entity) may be tasked only with administering or organizing the proceedings, not issuing the decision.

For domestic arbitration, Article 1450 is a direct and categorical natural-person requirement. An AI system—even if operated by or housed within a legal entity—cannot itself exercise the arbitral mission. AI may assist a human arbitrator, but it cannot alone render a domestic French arbitral award.

B. International Arbitration: No Verbatim “Personne Physique” Rule, Yet Similar Implications

For international arbitration, Article 1506 incorporates selected domestic provisions, and Article 1450 is not among those automatically incorporated. The international provisions therefore do not repeat the natural-person rule verbatim. That omission does not establish that an AI appointment is valid; the issue remains untested and must be read against the rest of the statutory scheme.

Articles 1456–1458, applied to international arbitration through Article 1506, address acceptance, disclosure, challenge, prevention, abstention, and resignation. They presuppose an arbitrator capable of exercising personal independence and impartiality and of being removed or replaced. Those provisions strongly support a human office-holder even though Article 1450 is not directly incorporated.

The safer conclusion is that French international arbitration law also contemplates an identifiable human arbitrator who remains responsible for the proceedings and award. A French court has not established a general rule validating an autonomous AI arbitrator, and such an arrangement would face substantial constitution, due-process, signature, and enforcement objections.

C. Possible Indirect Challenges and Public Policy

Even if parties somehow attempted to sidestep the requirement of a “personne physique,” the code’s enforcement provisions would pose other obstacles:

• Due Process (Principe de la Contradiction). Articles 1464 and 1510 require observance of equality and the adversarial principle. A tribunal that relies on material AI-derived facts, authorities, or theories without giving the parties a fair opportunity to address them may violate those requirements.
• Setting Aside or Refusal of Exequatur. Articles 1492 and 1520 provide limited grounds for annulment or refusal, including irregular constitution of the tribunal and failure to respect the adversarial principle. AI use matters only insofar as it produces one of those recognized defects.
• Manifest Contrariety to Public Policy. Articles 1488 and 1514 require refusal of enforcement where the award is manifestly contrary to domestic or international public policy, as applicable. A wholly automated process that eliminates the legally responsible tribunal or fundamentally compromises procedural fairness could attract that scrutiny.

The French framework therefore permits AI assistance but places the arbitral mission, procedural duties, and award responsibility on the legally appointed arbitrator or tribunal.

Netherlands

Under the Dutch Arbitration Act (Book 4 of the Dutch Code of Civil Procedure), Article 1023 expressly provides that every legally competent natural person may be appointed as an arbitrator. The statute therefore does more than merely imply a human decision-maker: it ties appointment to natural-person status and legal capacity.

Other provisions reinforce that structure. Arbitrators must disclose circumstances that may justify doubts about impartiality or independence and may be challenged; Article 1057 requires the award to be signed; and the tribunal exercises discretionary powers over procedure, evidence, and hearings. AI can support those functions, but the appointed natural person remains responsible for them.

Accordingly, an autonomous AI system cannot be appointed as the arbitrator under Article 1023. It may be used as an assistive tool, subject to the parties’ agreement, due process, confidentiality, data-protection duties, and the arbitrator’s independent judgment.

Sweden

Under the Swedish Arbitration Act (1999:116), Section 7 provides that a person who possesses full legal capacity in relation to themselves and their property may serve as an arbitrator. Read with the Act’s provisions on impartiality, disclosure, challenge, termination, and signature, the statutory office is directed to a legally capable individual rather than an autonomous system.

Sections 8 and 9 address impartiality and disclosure, and Section 31 requires the award to be in writing and signed by the arbitrators. The tribunal must also exercise procedural and decisional judgment for which the appointed arbitrators remain accountable. AI may assist, but it cannot replace those statutory responsibilities.

The Swedish Act does not use the phrase “AI is prohibited,” but its capacity, disclosure, challenge, and signature requirements make an autonomous non-human arbitrator incompatible with the present statutory scheme.

Spain

Under the Spanish Arbitration Act (Ley 60/2003, de 23 de diciembre), Article 13 expressly provides that natural persons in the full exercise of their civil rights may be arbitrators, subject to any professional restrictions imposed by applicable law. The requirement therefore directly excludes an autonomous AI system from appointment as arbitrator.

Articles 16–18 govern acceptance, independence, impartiality, disclosure, and challenge, and Article 37 requires the award to be signed by the arbitrator or arbitrators. AI may assist those individuals, but it cannot assume the statutory office or the responsibility for the award.

Italy

Under Articles 806–840 of the Italian Code of Civil Procedure, the statutory scheme likewise presupposes a legally capable, appointed individual exercising the arbitral office. For example:

• Article 812 excludes any person who lacks legal capacity to act from serving as arbitrator. An autonomous AI system cannot satisfy that personal capacity requirement.
• Articles 813 and 815 address acceptance, disclosure, and challenge, all of which attach to the appointed arbitrator and their personal independence and impartiality.
• Article 823 requires the award to be in writing and signed by the arbitrators; Article 824-bis addresses its effects; and Article 813-ter provides for arbitrator liability in specified circumstances. These provisions reinforce personal responsibility for the arbitral office.

Although the Code does not expressly mention AI, its legal-capacity, acceptance, challenge, signature, and liability rules effectively exclude an autonomous AI from serving as arbitrator. AI assistance remains possible under human control and subject to procedural fairness and confidentiality.

Other Jurisdictions

United Arab Emirates

Under Federal Law No. 6 of 2018 Concerning Arbitration, as amended, the UAE expressly requires an arbitrator to be a natural person with full legal capacity and other statutory qualifications. Specifically:

1. Article 10(1)(a) requires the arbitrator to be a natural person who is not a minor, under interdiction, or deprived of civil rights by reason of bankruptcy or specified criminal conviction. Article 10(1)(b) restricts service by members of the board of trustees or the administrative body of the institution administering the case, subject to the conditions later introduced in Article 10 bis.
2. Article 10(2) confirms that an arbitrator need not have a particular nationality or gender unless the parties or applicable law provide otherwise. That flexibility operates within the threshold requirement that the arbitrator be a natural person.
3. Article 10(3) requires a prospective arbitrator to disclose circumstances likely to cast doubt on impartiality or independence and to continue disclosing relevant circumstances throughout the proceedings. The duty is personal to the arbitrator.
4. Article 10 bis, added by amendment, permits a person drawn from certain supervisory or controlling bodies of an arbitral institution to serve only under specified conditions, including governance and disclosure safeguards. It is an exception governing eligible natural persons; it does not authorize a non-human arbitrator.

The UAE statute therefore expressly excludes an autonomous AI system from serving as the sole or true arbitrator. AI may be used as a tool, but the natural-person arbitrator must retain independence, impartiality, procedural control, and responsibility for the award.

Singapore

Under Singapore’s International Arbitration Act 1994 (2020 Revised Edition) (the IAA), which gives the UNCITRAL Model Law force of law with modifications, there is no express sentence stating that an arbitrator must be human. The appointment, disclosure, challenge, immunity, procedural-power, and signature provisions nevertheless strongly presuppose a legally accountable person or panel. No reported Singapore decision has validated an autonomous AI as arbitrator.

A. Terminology and Structure of the Legislation

Section 2 defines “arbitral tribunal” to include a sole arbitrator, a panel of arbitrators, a permanent arbitral institution, and an emergency arbitrator. Article 2 of the Model Law refers to a sole arbitrator or panel. The inclusion of a permanent arbitral institution does not establish that software may act as the merits decision-maker; the rest of the Act operates through appointed arbitrators and institutional functions.

The Act and Model Law consistently address an arbitrator who can be appointed, disclose circumstances, be challenged or replaced, exercise procedural powers, sign an award, and bear the legal consequences attached to the office.

B. Core Provisions That Imply a Natural Person Arbitrator

Appointment and Acceptance

Section 9A of the IAA, read with Article 11 of the Model Law, provides mechanisms for appointing the arbitrator where the parties’ agreed process does not produce an appointment. The scheme contemplates an identifiable appointee or panel capable of assuming the legal office.

Article 12(1) of the Model Law begins: “When a person is approached in connection with his possible appointment as an arbitrator….” That person must disclose circumstances likely to give rise to justifiable doubts as to impartiality or independence.

Disclosure of Potential Conflicts

Article 12 requires continuing disclosure and Article 13 permits challenge for justifiable doubts about impartiality or independence or for lack of agreed qualifications. Those duties are directed to the appointed arbitrator; technical disclosures by an AI provider do not substitute for them.

Signatures, Liability, and Immunities

Section 25 protects an arbitrator from liability for negligence in the capacity of arbitrator and for mistakes of law, fact, or procedure. Section 25A separately protects an appointing authority or arbitral institution unless it acts in bad faith. Article 31(1) of the Model Law requires the award to be in writing and signed by the arbitrator or arbitrators. These provisions attach legal responsibility and formal acts to the recognized office-holder.

Procedural Powers That Depend on Human Judgment

Section 12(1) confers procedural powers including security for costs, discovery, interrogatories, evidence preservation, and interim measures. Section 12(3) permits inquisitorial processes where the tribunal thinks fit. The substantive rules of law are addressed by Article 28 of the Model Law, not Section 12(5); Section 12(5) concerns the tribunal’s remedial powers. These functions may be AI-assisted, but the tribunal must independently exercise them.

C. Arbitrator’s Duties and the Necessity of Human Capacity

Many duties that the IAA imposes on arbitrators are inherently personal and judgment-based:

• Fair Hearing and Due Process. Articles 18 and 24 of the Model Law require equal treatment, a full opportunity to present the case, and fair management of hearings or documentary proceedings. AI use must not deprive a party of the opportunity to address material facts, authorities, or reasoning.
• Ability to Be Challenged, Replaced, or Terminated. Articles 12–15 establish procedures for challenge, termination, and substitution of arbitrators. The framework presupposes an identifiable office-holder to whom independence, impartiality, capacity, and mandate can be attributed.
• Signing and Delivering the Award. Article 31 requires a written award signed by the arbitrator or arbitrators and delivered to the parties. AI may assist drafting, but the appointed tribunal must review, adopt, sign, and take responsibility for the award.

D. Permanent Arbitral Institutions vs. Automated “AI Arbitrators”

Section 2’s reference to a “permanent arbitral institution” does not mean that an institution—or an AI system operated by it—automatically replaces the appointed tribunal as merits decision-maker. The Act distinguishes institutional functions from the adjudicative functions exercised by the arbitrator or panel:

1. The arbitral institution may administer proceedings, perform appointing functions, provide case-management services, or exercise other powers conferred by the rules and law.
2. The appointed arbitrator or panel conducts the adjudication, exercises the tribunal’s powers, reaches the decision, and signs the award.

Ethical and Due Process Considerations

The use of AI in arbitration gives rise to several ethical and due process concerns. Arbitration is founded on principles of fairness, party autonomy, and the right to be heard; introducing AI must not undermine these fundamentals. Key considerations include:

• Transparency and Disclosure: There is no universal rule requiring disclosure of every AI use. The appropriate approach is context-specific. Disclosure is particularly relevant where AI supplies material factual or legal analysis, introduces information outside the record, affects evidence, or is required by law, professional rules, institutional guidance, or a procedural order. The 2024 SVAMC Guidelines state that disclosure is not necessary as a general matter, while SCC, VIAC, Ciarb, and other guidance provides frameworks for consultation or disclosure in material circumstances.
• Bias and Fairness: AI systems can introduce or amplify bias through training data, design choices, deployment context, or feedback loops. In arbitration, a tool may underrepresent legal traditions, languages, industries, or classes of claimant. Representative data and testing help, but the central safeguard is independent human review, comparison against the record, and refusal to treat model output as authoritative.
• Due Process and Right to a Fair Hearing: Each party must have a fair opportunity to present its case and respond to the other side’s case. If the tribunal intends to rely materially on an AI-derived fact, authority, calculation, or theory that the parties have not addressed, it should consider whether the parties must be given an opportunity to comment. The tribunal’s reasons must be its own and must be capable of being stated in the award.
• Confidentiality and Privacy: Case materials should not be uploaded to a public or consumer AI service without understanding data retention, model-training, access, security, privilege, and cross-border transfer consequences. Appropriate safeguards may include enterprise terms, access controls, encryption, data minimization, redaction or anonymization, processor agreements, and incident-response procedures. Where personal data are processed in the EU context, the GDPR continues to apply independently of the AI Act, including rules on lawful basis, purpose limitation, security, processors, and international transfers.
• Professional Responsibility of Counsel: Lawyers remain responsible for work performed with generative AI. ABA Formal Opinion 512 (July 29, 2024), while based on the ABA Model Rules and not itself binding in every jurisdiction, identifies duties of competence, confidentiality, communication, supervision, candor, meritorious advocacy, and reasonable fees. Counsel should independently verify AI-generated authorities and factual assertions and should not submit output that has not been checked against the record and official sources.

AI Use Cases and Real-World Examples

Despite the concerns, AI is already making tangible contributions in arbitration practice. A number of use cases and real-world examples demonstrate how AI tools are being applied by arbitral institutions, arbitrators, and parties:

Document Review and E-Discovery: Arbitration cases, especially international commercial and investment disputes, often involve large document collections. Technology-assisted review, clustering, semantic search, and machine-learning classification can accelerate review and help identify potentially relevant material. Platforms such as Relativity incorporate these functions, but sampling, validation, privilege controls, and human quality assurance remain necessary.

Language Translation and Interpretation: Machine translation, speech recognition, and real-time transcription can assist in multilingual proceedings and reduce cost and delay. They should not be treated as authoritative for contested testimony, operative contract language, or award text without qualified human review, because errors can alter legal meaning or evidentiary nuance.

Legal Research and Case Analytics: AI-assisted legal research can retrieve authorities, summarize materials, and support citation analysis. CoCounsel is now a Thomson Reuters product, and arbitration-focused databases and institutional libraries increasingly incorporate AI search. Every proposition, quotation, citation, and procedural rule must still be checked against the official source; fluent output is not proof of legal accuracy.

Arbitrator Selection and Conflict Checking: Data tools can organize public information about experience, appointments, publications, and disclosed relationships, and can support broader conflict searches. They do not replace the candidate’s continuing disclosure duty, institutional review, or counsel’s independent investigation, and they may reproduce gaps or biases in the underlying data.

Case Management and Procedural Efficiency: Arbitral institutions have begun integrating AI to streamline case administration. Furthermore, during proceedings, AI chatbots can answer parties’ routine questions about rules or schedules, easing the administrative burden.

Another emerging use is AI-supported settlement analysis. A party may use scenario modeling, probability ranges, or damages simulations as one input when evaluating settlement. Such output should be treated as decision support, not an objective forecast: arbitration data are often confidential and incomplete, and model assumptions may be opaque or poorly matched to the dispute.

Arbitration Institutions Implementing AI Initiatives

ICC (International Chamber of Commerce)

The ICC’s “Overarching Narrative on Artificial Intelligence” is a general business-policy document rather than an arbitration rule. More directly, the ICC Commission on Arbitration and ADR announced a Task Force on Artificial Intelligence in International Dispute Resolution in September 2024 to develop practical guidance while protecting fundamental principles of international dispute resolution. In January 2026, the ICC Dispute Resolution Library became accessible through Jus Mundi’s Jus AI research tool. These initiatives support research and governance; they do not authorize delegation of the tribunal’s adjudicative responsibility.

AAA-ICDR (American Arbitration Association & International Centre for Dispute Resolution)

The AAA-ICDR published the AAAi Standards for Use of AI in ADR in May 2025, supplementing its guidance for arbitrators and other participants. Its tools include AI-assisted transcription and the AAAi Panelist Search. In November 2025, the AAA launched an “AI Arbitrator” workflow for specified two-party, documents-only construction disputes. Despite the name, the system analyzes submissions, surfaces issues, and prepares a proposed award; a human arbitrator reviews the record, makes the final decision, revises as necessary, signs, and issues the award.

JAMS (USA)

JAMS introduced its Artificial Intelligence Disputes Clause and Rules, effective June 14, 2024, for disputes involving AI systems. JAMS expressly distinguishes those case rules from rules governing a neutral’s or advocate’s use of AI in the dispute-resolution process. Later in 2024, JAMS launched JAMS Next, including AI-assisted transcription and natural-language neutral search. Those operational tools remain subject to human neutral responsibility and JAMS governance.

SCC (Arbitration Institute of the Stockholm Chamber of Commerce)

On October 17, 2024, the SCC released its non-binding Guide to the use of artificial intelligence in cases administered under the SCC Rules. The Guide addresses confidentiality, competence, verification, transparency, and non-delegation. It encourages disclosure by a tribunal where AI is used in researching or applying facts and law in a way relevant to the decision, while preserving case-specific flexibility.

CIETAC (China International Economic and Trade Arbitration Commission)

CIETAC has continued to digitize case administration and online dispute resolution. Its 2024 Work Report describes expanded digital case services, and its 2025 Work Report states that CIETAC will actively explore deeper integration of artificial intelligence and big data with arbitration. Those statements concern institutional modernization and service delivery; they do not indicate that AI replaces appointed arbitrators.

Silicon Valley Arbitration & Mediation Center

The SVAMC Guidelines on the Use of Artificial Intelligence in Arbitration (1st Edition, published April 30, 2024) provide a principle-based framework for arbitrators, parties, counsel, and experts. They address competence, confidentiality, verification, hallucinations, bias, evidence integrity, and non-delegation. Guideline 3 states that disclosure is not required as a general matter and should be assessed case by case, including due-process and privilege considerations. The Guidelines include a model clause for procedural orders and remain designed to evolve with the technology.

Chartered Institute of Arbitrators (Ciarb)

Ciarb launched its Guideline on the Use of AI in Arbitration in March 2025 and issued an updated version in September 2025. The Guideline addresses benefits and risks, general recommendations for participants, tribunal powers to regulate party use, and arbitrator use of AI. It includes a template agreement and template procedural order, and it emphasizes consultation, verification, confidentiality, procedural fairness, and the tribunal’s non-delegable responsibility for decisions.

Vienna International Arbitral Centre (VIAC)

VIAC published its non-binding Note on the Use of Artificial Intelligence in Arbitration Proceedings in April 2025. The Note is intended to facilitate discussion in VIAC cases and addresses competence, responsibility for outputs, confidentiality, data security, evidence, disclosure, and tribunal management. VIAC does not endorse a particular tool, and participants remain responsible for compliance with applicable law and the Vienna Rules.

Practical Takeaways

AI can improve arbitration when its role is bounded, verified, secure, and auditable. Parties and tribunals should address material AI use at the first procedural conference; define any prohibited or disclosure-triggering uses; protect confidential and personal data; verify citations, calculations, translations, and summaries; preserve a reliable record of material inputs and outputs where appropriate; and ensure that no participant treats model output as authoritative.

Autonomous AI arbitrators remain expressly excluded in several jurisdictions and legally doubtful in others. The presently defensible model is AI-assisted arbitration: an identifiable, independent human arbitrator or tribunal controls procedure, evaluates the evidence and law, reaches the conclusions, provides the reasons, signs the award, and accepts responsibility for the result.