Artificial Intelligence In Arbitration: A Necessary Evil Or a Threat To Due Process?

Abstract

Artificial intelligence has moved from the periphery of arbitral practice to its operational core, assisting tribunals and counsel with research, document review, translation and drafting. This essay asks whether that shift is a necessary evil arbitration must absorb to retain its advantages in speed and cost, or a genuine threat to the due process guarantees that legitimate arbitral awards. Using a doctrinal methodology, it examines the conceptual foundations of arbitration and AI, evaluates AI’s benefits, and interrogates the due process risks of opacity, algorithmic bias, breach of confidentiality and improper delegation of adjudicative authority. It reviews the regulatory soft-law landscape, including the SVAMC Guidelines, the CIArb Guideline, the AAA–ICDR Principles and Nigeria’s Arbitration and Mediation Act 2023, against recent disputes including LaPaglia v Valve Corp and PEN Guild v POLITICO. It argues that AI is best understood as a necessary evil whose risks are presently manageable only where its use is disclosed, supervised, and governed by enforceable rather than aspirational standards, and concludes with recommendations for binding regulation in Nigeria and beyond.

Keywords: Artificial Intelligence, Arbitration, Due Process, Integrity, Arbitral Awards, Arbitration and Mediation Act, SVAMC Guidelines, Nigeria.

1. Introduction

James submitted his dispute to an arbitration panel to avoid the bureaucratic delays of a court judgment. To his surprise, the panel delivered a questionable award against him, reportedly drafted with the assistance of ChatGPT, with the arbitrator keen to conclude matters before a planned summer holiday to the Galapagos Islands. Though illustrative, this scenario mirrors an actual challenge litigated before the courts of the United States and forms the empirical anchor for this essay’s discussion.[1]

The growth of arbitration as the preferred mechanism for resolving commercial disputes has coincided with the rapid diffusion of artificial intelligence (“AI”) across the legal profession. Arbitrators, counsel and institutions now routinely deploy AI for legal research, document review, translation, transcription and drafting support. Yet the same features that make AI attractive, namely speed and the capacity to manage voluminous data at lower cost, raise acute due process concerns: opacity, the absence of meaningful avenues of challenge, and the risk that an arbitrator’s non-delegable adjudicative function is silently displaced.

This essay addresses a discrete question: does the unregulated use of AI in arbitration constitute a “necessary evil” the system must tolerate to preserve its efficiency and party autonomy, or has it crossed into a genuine threat to due process, namely the right to a fair hearing and the principle of audi alteram partem, which legitimates arbitral awards. The essay argues that AI is, at present, a necessary evil rather than an irredeemable threat, but only conditionally so: its risks are tolerable only with disclosure, human oversight and binding rather than aspirational regulation. Absent these safeguards, AI use tips from convenience into a structural threat to legitimacy.

The essay proceeds in eight further parts: the conceptual framework of arbitration and AI, the case for AI as a tool, the due process risks, regulatory responses, recent arbitral disputes, critical analysis, recommendations and a conclusion.

2. Conceptual Framework: Arbitration and Artificial Intelligence

2.1 The Nature and Rationale of Arbitration
The World Intellectual Property Organization defines arbitration as a procedure in which a dispute is submitted, by agreement of the parties, to one or more arbitrators who render a binding decision. In choosing arbitration, parties opt for a private dispute resolution mechanism instead of recourse to the courts. Arbitration is consensual, neutral and confidential; the parties select the arbitrator(s), and the award is binding and enforceable save where fraud, breach of due process or some other vitiating element is established. Arbitration also offers technical expertise not always readily available in the ordinary courts.

Arbitration has proved valuable where parties desire swift settlement, where the dispute calls for expertise the courts do not readily possess, or where ordinary litigation would be unduly costly. It has been described as the future of dispute settlement, a description borne out by its recognition in statute across multiple jurisdictions. Yet arbitration is not flawless. Proceedings tainted by fraud can undermine it, as the English Commercial Court found in The Federal Republic of Nigeria v Process & Industrial Developments Ltd,[2] where an eleven-billion-dollar award was set aside for bribery and perjured evidence. An award may also prove practically toothless where a court order is needed for enforcement, as in Red Lion Hotels Franchising, Inc v Leslie,[3] decided under the US Federal Arbitration Act 1925.

2.2 The Rise of Artificial Intelligence in Legal Practice
Technological advancement in dispute resolution was inevitable. The first documented AI systems date to the early 1950s, and AI has since become ubiquitous worldwide, a trend accelerated by generative systems such as ChatGPT, which reportedly reached several hundred million weekly active users within a few years of launch. Legal professionals are increasingly reliant on AI for tasks that once called for independent cognitive judgment. The Chief Justice of Nigeria addressed this trend at a homecoming lecture at the University of Lagos in June 2025, while a co-panelist cautioned against unsupervised use of AI in legal research and writing. AI use is now commonplace in legal research, writing and proceedings, including among law students, more than half of whom report using it for academic assignments. The question, then, is what role AI should play in arbitration, and at what cost?

3. The Case for AI in Arbitration: A Necessary Tool

AI has been credited with enhancing efficiency, lowering costs and streamlining complex arbitral processes, functioning principally as a tool for analysis, document management and drafting support. It has been deployed for legal research, evidence review, document translation, and real-time transcription of hearings. Institutions have begun building AI into their architecture: the American Arbitration Association’s International Centre for Dispute Resolution (AAA–ICDR) started using AI in October 2024 to help case managers select suitable arbitrators, while HKIAC now generates AI summaries of its procedural decisions. The International Chamber of Commerce’s (ICC’s) 2022 Report on Information Technology in International Arbitration found that most surveyed practitioners considered such tools to have improved efficiency.

These gains matter because arbitration’s main advantage over litigation is speed and economy. Where AI genuinely reduces time and cost without compromising process integrity, it serves arbitration’s foundational rationale. The difficulty, examined below, is that the same capacity to synthesise information without a transparent chain of reasoning is what generates due process risk.

4. The Case Against AI in Arbitration: A Threat to Due Process

4.1 The “Black-Box” Problem and the Right to be Heard
The most common objection to AI in adjudicative settings is the so-called “black-box” problem: even sophisticated machine-learning models cannot render their reasoning intelligible to the humans who rely on their output. Where an arbitrator relies, even partially, on an undisclosed AI-generated analysis, the right to be heard, recognised under Article V(1)(b) of the New York Convention 1958,[4] may be compromised: a party cannot know whether the award rests on the record before the tribunal or on extrinsic material it never had the chance to contest.

This concern echoes the common law’s insistence, articulated by Lord Denning in Ridge v Baldwin,[5] that no person be condemned unheard on grounds they never had the opportunity to address. Where an AI system assesses witness credibility or ranks submissions through opaque processes, an affected party has no meaningful way to interrogate its errors. Scholarship on algorithmic justice describes this as a procedural gap: fairness tools aimed at bias address distributive justice, while the distinct demand of “voice,” namely consent, transparency and a genuine right to be heard, is comparatively neglected.

4.2 Algorithmic Bias and Structural Injustice
A second risk is algorithmic bias. Where training data over-represents particular categories of contract, jurisdiction or legal tradition, AI outputs tend to favour those categories, entrenching rather than correcting structural injustice; a model trained mainly on awards from Western-seated tribunals may, for instance, undervalue claims framed in civil-law terms. An AI layer trained on historically skewed data risks compounding, rather than correcting, arbitration’s existing diversity imbalance.

4.3 Confidentiality and Data Protection Risks
Confidentiality is one of arbitration’s principal advantages over litigation. Generative AI tools complicate this advantage, since feeding case materials into a third-party system may amount to unauthorised disclosure, exposing arbitrators and counsel to liability for breach of confidentiality. Commentators have documented instances where law firms using cloud-hosted generative AI inadvertently exposed client data through the platforms’ terms of service.

4.4 Delegation of Decision-Making Authority
The most fundamental objection is structural: arbitration rests on the parties’ consent to have their dispute decided by the human arbitrator(s) they chose, not an undisclosed algorithm. The principle delegatus non potest delegare is reflected in section 48(1)(a)(iv) of the Arbitration and Mediation Act 2023,[6] which permits an award to be set aside where the tribunal’s composition or procedure departed from the parties’ agreement. The ICC’s Note to Parties and Arbitral Tribunals draws a useful analogy with arbitral secretaries: a tribunal may use a secretary, or by extension an AI tool, for research or drafting, but may never delegate its decision-making function. Where an arbitrator allows AI to perform analytical work properly belonging to the human adjudicator, the resulting award is vulnerable as the product of a tribunal that never, in the relevant sense, decided the case.

5. Regulatory and Institutional Responses

5.1 International Soft-Law Instruments
Because no state has yet enacted binding legislation on AI use in arbitration, the field is governed by an expanding patchwork of institutional soft law. The Silicon Valley Arbitration and Mediation Center (“SVAMC”) published the first dedicated instrument, its Guidelines on the Use of Artificial Intelligence in Arbitration, on 30 April 2024.[7] The Guidelines impose a “human-in-the-loop” requirement, meaning a human must remain responsible for the final, signed award; require disclosure of AI use; guard against AI-generated false evidence; and require attorneys to verify AI output to avoid liability for “hallucinated” authorities. Other institutions followed quickly: the AAA–ICDR published Principles Supporting the Use of AI in Alternative Dispute Resolution in 2023, without an explicit disclosure obligation; the Stockholm Chamber of Commerce published a Guide encouraging, though not mandating, disclosure in October 2024; the CIArb published its own Guideline in 2025; and the Vienna International Arbitral Centre issued a Note in April 2025 leaving disclosure to the tribunal’s discretion. In September 2024 the ICC Commission on Arbitration and ADR announced a dedicated Task Force on Artificial Intelligence in International Dispute Resolution. Notably, the rules of major institutions, including the ICC, LCIA, SIAC and HKIAC, remain silent on AI use, leaving soft law to fill a gap institutional rules have not yet addressed.

5.2 The Nigerian Position
Nigeria has no statute regulating AI use in arbitration specifically. Two instruments are nonetheless relevant: the Nigerian Bar Association Section on Legal Practice’s 2024 Guidelines for the Use of Artificial Intelligence in the Legal Profession in Nigeria, which address lawyers’ competence, confidentiality and supervision obligations but are professional-conduct instruments, not arbitration-specific, and the Arbitration and Mediation Act 2023, whose provisions on transparency, arbitrator impartiality and grounds for setting aside an award may be expansively interpreted to capture improper AI use. This is an interpretive stretch rather than deliberate design, and the gap between rapid AI uptake and the absence of purpose-built regulation is, in this essay’s argument, the central deficiency requiring legislative attention.

6. Case Analysis: Lessons from Recent Arbitral Disputes

6.1 LaPaglia v Valve Corp
In LaPaglia v Valve Corp,[1] James LaPaglia initiated an arbitration against Valve Corporation alleging anticompetitive practices and exclusion from its Steam marketplace. The sole arbitrator issued a twenty-nine-page final award for Valve just fifteen days after final briefing. LaPaglia petitioned to set the award aside on the ground that the arbitrator had “outsourced his adjudicative role to Artificial Intelligence,” pointing to the arbitrator’s disclosure that he had previously used ChatGPT to draft an article, and his stated wish to conclude the matter before a holiday to the Galapagos Islands. On the facts pleaded, the allegation implicates both the “human-in-the-loop” and procedural integrity requirements of the SVAMC Guidelines: an opaque award produced with unusual speed raises a legitimate question whether the arbitrator performed the function the parties bargained for. Its resolution is likely to remain a landmark reference point on the threshold at which AI-assisted drafting tips into impermissible delegation.

6.2 PEN Guild v POLITICO
In PEN Guild v POLITICO,[8] the claimants arbitrated against POLITICO’s management over its unilateral introduction of AI tools that bypassed safeguards in a collective bargaining agreement and undermined journalistic standards. The arbitrator found POLITICO had violated the agreement by failing to give required notice, to bargain over the change, or to maintain human oversight of the AI tools, and ruled for the claimants. Though the case did not involve AI use by the tribunal itself, it illustrates a parallel point: AI deployed without consent, disclosure or human oversight will be treated as inconsistent with the procedural protections an agreement was designed to secure.

6.3 P&ID v Nigeria: A Cautionary Parallel
The Federal Republic of Nigeria v Process & Industrial Developments Ltd, though it predates the AI controversy, remains instructive by analogy. The English Commercial Court set aside an eleven-billion-dollar award after finding the process had been corrupted by bribery, perjured evidence and the claimant’s improper retention of Nigeria’s privileged legal documents through a corrupted official. The case stands for the broader proposition that English, and by extension Nigerian, courts will set aside an award where the integrity of the process is compromised, the same willingness to scrutinise process over outcome that claimants in LaPaglia invoke, and that future Nigerian claimants are likely to invoke where an award is suspected of resulting from undisclosed or unsupervised AI use.

7. Critical Analysis and Original Viewpoint

AI in arbitration is best characterised neither as an unqualified good nor an irredeemable evil, but as a necessary evil whose risks are presently manageable, provided three conditions are met: disclosure, human oversight and binding regulation. Where any one is absent, AI use ceases to be a tolerable cost of efficiency and becomes a genuine threat to due process.

The weakness common to nearly every soft-law instrument surveyed above is enforceability. The SVAMC Guidelines, the CIArb Guideline and the AAA–ICDR Principles articulate best practice, but none carries the force of law; a tribunal that ignores them breaches no binding rule. The consequence falls on the party least able to detect it, as LaPaglia demonstrates: a party facing an opaque, fifteen-day award has no practical means of establishing AI involvement beyond the arbitrator’s own voluntary disclosure. This asymmetry is, in the writer’s view, the most significant due process defect in the current regulatory architecture.

The distinction between AI as a “tool” and AI as a “decision-maker” is, in practice, less stable than it appears. The ICC’s arbitral-secretary analogy assumes a bright line between permissible drafting assistance and impermissible decision-making. Yet where AI summarises voluminous evidence, ranks submissions, or drafts a substantial portion of an award’s reasoning, the line between assistance and substitution becomes a matter of degree, leaving a binary regulatory distinction inadequate to the risk it seeks to manage.

8. Recommendations

Although AI may prove useful in arbitration, its use must be bindingly regulated. First, Nigeria and other jurisdictions presently relying on soft law should enact binding statutory provisions addressing AI use in arbitration, drawing on the SVAMC Guidelines and the CIArb Guideline; a targeted amendment to the Arbitration and Mediation Act 2023, inserting express disclosure and human-oversight obligations, would be a proportionate first step. Second, arbitrators must take care in inputting confidential materials into AI systems, given the risk of breaching their duty of confidentiality under section 57 of the Act.

Third, the parties’ informed consent should be sought before AI is used in any substantive aspect of the proceedings, recorded in the terms of reference. Fourth, where AI use is permitted, its output must be subjected to rigorous human verification before incorporation into any order or award. Fifth, courts seised of a setting-aside or enforcement application should set aside an award shown to result from improper or non-consensual AI use, treating this as a breach of due process within section 36 of the Constitution of the Federal Republic of Nigeria 1999.[9] This is the judicious performance of the judiciary’s role as guardian of procedural fairness.

9. Conclusion

Arbitration remains an efficient and increasingly indispensable means of resolving commercial disputes, offering a quicker, more technically informed alternative to courts with heavy caseloads. AI has materially affected arbitral practice, offering genuine efficiency gains in research, document review, translation and case administration, while raising serious due process concerns: opacity and the erosion of the right to be heard, algorithmic bias, breaches of confidentiality, and the risk that an arbitrator’s non-delegable decision-making function is silently displaced.

This essay has argued that AI use in arbitration is best understood as a necessary evil rather than an outright threat to due process, but only where it is disclosed, supervised by a human decision-maker, and governed by binding rather than aspirational standards. Where these conditions are absent, as the dispute in LaPaglia v Valve Corp illustrates, the line between efficient tool and due process violation is crossed. The recommendations above, namely statutory reform, mandatory disclosure and consent, rigorous human verification, and judicial willingness to set aside improperly AI-assisted awards, are offered as the means by which Nigeria and the wider arbitral community might capture AI’s benefits while containing its risks to legitimacy.

References

[1] LaPaglia v Valve Corp, Petition to Vacate Arbitration Award, No. 3:25-cv-00833 (S.D. Cal., filed 8 Apr. 2025).

[2] The Federal Republic of Nigeria v Process & Industrial Developments Ltd [2023] EWHC 2638 (Comm).

[3] Red Lion Hotels Franchising, Inc v Leslie, decided under the Federal Arbitration Act, 9 U.S.C. §§ 1 et seq. (1925).

[4] Convention on the Recognition and Enforcement of Foreign Arbitral Awards (New York Convention), art. V(1)(b), June 10, 1958, 330 U.N.T.S. 38.

[5] Ridge v Baldwin [1964] AC 40 (HL).

[6] Arbitration and Mediation Act 2023 (Nigeria), s. 48(1)(a)(iv).

[7] Silicon Valley Arbitration & Mediation Center, Guidelines on the Use of Artificial Intelligence in Arbitration (30 Apr. 2024).

[8] PEN Guild v POLITICO, Arbitration Award (26 Nov. 2025).

[9] Constitution of the Federal Republic of Nigeria 1999, s. 36.