Author: Qerim Qerimi*
| Jurisdiction: | Topics: |
I. Introduction
Imagine a case where every participant in the proceeding, attorneys preparing submissions, experts producing their reports, and even the tribunal drafting its award, relies on Artificial Intelligence (AI) as an integral part of their work. What was once imaginary is now becoming routine. The unprecedented integration of AI into the architecture of contemporary dispute resolution has prompted a reconsideration of foundational principles across legal systems and domains. In domestic and international courts, as well as arbitral tribunals, AI systems now increasingly shape the approach about how facts are analyzed, arguments formulated, evidence assessed, and decisions drafted. AI tools already perform institutional tasks that range from document-review and pattern detection to advanced language models, automated transcription, and decision-support analytics or generative reasoning models.
While these systems promise increased efficiency and cost-effectiveness, they also raise questions central to arbitral legitimacy and fairness: transparency, accountability, equality and non-discrimination, data protection, and the human character of adjudication.
The Council of Europe’s European Commission for Democracy through Law (the Venice Commission) has addressed these issues with particular nuance in its Updated Rule of Law Checklist, introducing a new section on digital technologies and AI within the judicial system. Unlike domestic courts, international arbitration lacks a uniform public regulatory framework. It is governed by a combination of national arbitration laws, soft-law instruments, and the procedural rules of arbitral institutions. In this context, the regulatory lacuna makes the Venice Commission’s Rule of Law Checklist uniquely relevant. Although the Checklist is formally directed at state judicial systems, many of its principles have clear relevance for international arbitration, an adjudicatory mechanism that, despite its private nature, performs a quasi-public function in resolving cross-border disputes. Indeed, whereas international commercial arbitration exemplifies a private, consent-based model of transnational adjudication, albeit one that depends on state courts for enforcement and procedural support, investor-state arbitration blends the private form of arbitration with public-law substance, subjecting sovereign regulatory choices to treaty-based review and thus functioning as a hybrid institution within global governance.
This commentary argues that the Venice Commission’s standards should apply mutatis mutandis to international arbitration, providing a principled framework for responsible AI integration. Before examining how the Checklist maps onto arbitral practice, this commentary briefly surveys the positions of leading institutions and authorities on the use of AI in international arbitration.
II. Institutional and Doctrinal Positions on AI in Arbitration
A. United Nations Commission on International Trade Law (UNCITRAL)
International standard-setting organizations have begun articulating principles for AI governance in dispute resolution. UNCITRAL’s Working Group II has initiated consultations on technology-related dispute settlement and adjudication, emphasizing due process rights. It noted the advantages of such technology to manage cases more efficiently and to offer parties innovative methods to present and display evidence, while at the same time observing that disparities in access to technology between parties could undermine “the fairness of the proceedings,” requiring corrective measures (¶ 17).
B. International Centre for Settlement of Investment Disputes (ICSID)
No ICSID rule amendments expressly regulate AI, but the new 2022 ICSID Rules emphasize transparency, disclosure obligations, and data protection, principles directly relevant for AI adoption and governance.
In practice, ICSID tribunals have already taken positions and clarified the stance over the use of AI. In one of its procedural orders in ICSID Case No. ARB/24/46, an ICSID tribunal has adopted and disclosed the policy of the use of AI tools to perform such tasks as “preparation of summaries, searching and organizing the record and legal research” (¶ 28.1). Yet the tribunal notes that it “will not delegate any decision-making functions to such AI tools,” and that “The Tribunal shall ensure that the AI tools that it uses provide adequate guarantees of cybersecurity and confidentiality” (¶ 28.2).
C. Practice of Arbitral Centers
Several arbitral institutions have issued preliminary guidelines or model administrative practices concerning AI.
International Chamber of Commerce (ICC) has incorporated references to AI tools and digital case management into its 2021 ICC report, entitled “Leveraging Technology for Fair, Effective and Efficient International Arbitration Proceedings.” The report discusses the use of machine learning AI, specifically predictive coding, in arbitration to identify relevant documents, offering potential time and cost savings, along with increased accuracy and transparency. However, it highlights the need for disclosure of predictive coding parameters to other parties and tribunals, as well as caution around AI-generated translations, which, while cost-effective, may raise concerns over accuracy and confidentiality, and should be used with appropriate safeguards (¶ 5.2).
American Arbitration Association – International Centre for Dispute Resolution (AAA-ICDR) issued in March 2025 a formal Guidance on Arbitrators’ Use of AI Tools. The Guidance sets out four core principles governing the appropriate use of AI tools by arbitrators: (1) ensuring the accuracy and reliability of AI-generated information; (2) maintaining fairness and due process; (3) preserving independent decision-making; and (4) being transparent with the parties. Substantively, apart from “maintaining the principles of fairness and due process” (¶ 2), arbitrators are required “to retain complete control over decision-making” (¶ 3). In context, and based on the content of the relevant paragraph, “complete control” entails the arbitrators’ continued exercise of their own judgment and expertise, including independent evaluation and reasoning.
Vienna International Arbitration Center (VIAC) has adopted a non-binding Note on the use of AI. VIAC recognizes the potential of AI to enhance the efficiency and accessibility of arbitration proceedings. It supports the integration of AI tools in areas such as legal research, document review, and case management, while emphasizing the need for human oversight and adherence to principles of fairness, transparency, and due process. The Note underscores that “Arbitrators retain full control over the decision-making process” (¶ 2.1).
D. Doctrinal Perspectives on the Use of AI in International Arbitration
Scholars and commentators in international arbitration agree on a number of foundational premises regarding AI’s potential and perils. While AI can accelerate the processing of complex cases (particularly in tasks such as research, document review, and procedural support), its outputs are fallible and require human oversight, especially as the obligation for reasoned decisions “is likely to be an important barrier for AI-based decisionmaking.”
Commentators have also highlighted the particular problem of opacity in AI outputs (the “black box” phenomenon), making transparency and traceability problematic, which in turn are essential to preserve due process and the enforceability of arbitral awards.
As Professor Sunstein observes in his recent book Imperfect Oracle: What AI Can and Cannot Do, AI may aid decision-making by detecting patterns and eliminating some human errors, but it remains fundamentally limited when outcomes depend on social, emotional, or cultural factors, and, in any event, “AI cannot always make accurate predictions,” underscoring the enduring role of human judgment.
The prevailing doctrinal view is that AI holds significant potential to enhance dispute resolution, but only when deployed within substantive and procedural safeguards that preserve fairness, transparency, and human responsibility. These concerns are precisely those operationalized by the Venice Commission in its updated Rule of Law Checklist.
III. Applying the Venice Commission’s AI Principles to International Arbitration
The Venice Commission’s updated Rule of Law Checklist, in particular its entirely new section on the use of AI within the judiciary, provides detailed standards for assessing the legitimacy of AI use within judicial institutions. Applied mutatis mutandis to international arbitration, albeit with due attention to relevant institutional and procedural nuances, these principles offer practical guidance for arbitral institutions, tribunals, and parties using AI tools. Accordingly, what follows is a proposed framework of core principles to govern AI use in international arbitration.
A. Legality, Transparency and Governance
The Rule of Law Checklist (¶ 125) underscores that while AI tools can enhance efficiency and access to justice, they also risk diminishing the right to a fair hearing, undermining privacy, and creating technological inequality. Consequently, their use must be governed by clear legal bases, transparent frameworks, and continuous human oversight.
In arbitration, however, “legality” must be understood within a pluralistic regulatory ecosystem. In practice, it means that arbitral institutions should adopt explicit AI governance policies, requiring disclosure of AI use, preserving confidentiality, and ensuring human oversight.
There are at least three principal ways how this demand can be effectuated in practical terms. One option is through procedural orders on AI use. Tribunals can incorporate AI-specific provisions addressing admissible tools, disclosure obligations, and data-security guarantees. The second option is through institutional AI policies, with institutions such as the PCA, ICSID, and ICC codifying standards in their Rules, requiring traceability and human oversight of AI-based services. The third option relies on party autonomy, allowing parties to negotiate AI-use clauses in their arbitration agreements that define permissible applications of AI and outline corresponding disclosure requirements.
These measures approximate the transparency and governance expectations articulated in the Rule of Law Checklist.
B. Explainability and Independent Oversight
The Checklist (¶ 126) requires that AI systems used in adjudication be transparent, explainable, and subject to independent oversight. For arbitration, the implications are especially significant for award enforcement. An award drafted or substantively shaped by AI without human-verifiable reasoning risks annulment or non-enforcement under the New York Convention on the Recognition and Enforcement of Foreign Arbitral Awards (1958), particularly under Article V(2)(b) (“The recognition or enforcement of the award would be contrary to the public policy of that country.”). To date, however, there appears to be no reported arbitral award whose annulment or non-enforcement has turned on the tribunal’s use of AI. Accordingly, the “public policy” concerns discussed here remain largely prospective, although they are likely to become increasingly significant as AI tools become more deeply and systematically integrated into arbitral practice.
The requirement that arbitral tribunals provide reasoned decisions poses a fundamental obstacle to the use of opaque or non-explainable AI systems, particularly given AI’s “black box” problem. Unless the parties expressly agree to dispense with reasons, arbitration frameworks impose a clear duty to articulate the basis of the decision, something that an unexplainable AI model cannot independently satisfy.
This obligation is clearly embedded in the UNCITRAL instruments: Article 31(2) of the UNCITRAL Model Law on International Commercial Arbitration stipulates that an award must state the reasons upon which it is based, save where the parties agree otherwise or the award records agreed terms, while Article 34(3) of the UNCITRAL Arbitration Rulesmirrors this requirement, providing that “[t]he arbitral tribunal shall state the reasons upon which the award is based, unless the parties have agreed that no reasons are to be given.” Comparable standards appear in domestic legal systems and in the rules of major arbitral institutions. For example, Section 52(4) of the English Arbitration Act 1996 mandates that an award include its reasons unless the parties have consented to omit them, and Article 32(2) of the ICC Ruleslikewise obliges tribunals to explain the reasoning underlying their awards (“The award shall state the reasons upon which it is based.”).
These provisions thus underscore that explainability is not optional but inherent to the adjudicative function in international arbitration. For this reason, arbitrators should reveal whether AI played a role in drafting the award or analyzing the record, without necessarily disclosing proprietary model details. At the same time, any disclosure obligation must be reconciled with arbitration’s traditional commitment to confidentiality. A balanced or proportionate approach would require transparency regarding the nature and extent of AI assistance sufficient to permit meaningful scrutiny, while avoiding disclosure of confidential case materials, deliberations, or proprietary technical information.
When it comes to oversight by institutions, independent oversight in the case of international arbitration may be operationalized through institutional review bodies assessing high-risk AI tools offered as part of institutional case-management systems.
These mechanisms help operationalize the Checklist’s demand for public trust and accountability.
C. Distinguishing Administrative and Adjudicative Functions
The Checklist (¶ 127) draws a fundamental distinction between the two underlying functions of any arbitral or judicial proceeding, namely administrative and adjudicative functions. In this context, AI used for administrative functions may follow general public-administration standards, while AI used in adjudicative functions requires heightened safeguards to protect fair-trial rights.
Applied to international arbitration, this distinction suggests two categories of AI tools: (1) low-risk administrative tools, and (2) high-risk adjudicative tools. Some AI applications, however, occupy an intermediate position. Predictive coding-systems used to identify potentially relevant documents, for instance, contribute to evidentiary process without themselves determining relevance, admissibility, or probative value. Their classification should therefore depend not on the technology itself, but on whether ultimate evaluative judgments remain under the independent control of the tribunal. The decisive criterion is thus not whether AI assists in the processing of evidence, but whether the tribunal retains the evaluative and decisional function. Against this background, the distinction between administrative and adjudicative uses of AI becomes clearer. AI tools used for translation, transcription, scheduling, and document sorting may be permissible so long as they comply with data protection norms and standards. On the other hand, predictive analytics, generative drafting tools, or AI systems that evaluate the weight of evidence should require explicit disclosure, party consent, and rigorous human review.
If AI is used to assist in legal research or draft sections of an arbitral award, arbitrators must ensure its contributions are verifiable and remain fully under the tribunal’s interpretive control. Ultimately, this distinction is indispensable to maintaining fairness and reliability of arbitral decisions.
D. The Human Decisional Element
Paragraph 128 of the Checklist articulates a bright-line rule: AI shall not replace human adjudication. Judicial decisions concerning fact-finding and legal application must remain the exclusive responsibility of judges and arbitrators.
The same principle applies, if not arguably more strongly, in international arbitration. A fully automated award, as now indicated, would likely violate public policy norms under most arbitration laws, and its enforceability would be doubtful under the New York Convention.
The implication of this principle is that arbitrators must be the genuine authors of the award, even if AI assists them in drafting. The reasoning must reflect the tribunal’s independent judgment. This same principle imposes an additional duty of verification, so that arbitrators must verify the accuracy of AI-generated citations, factual summaries, and legal propositions, as well as disclose where material assistance occurred. Training modules may need to be developed by arbitral institutions to ensure arbitrators understand AI’s capabilities and limitations.
This human-centered model is key to preserve the legitimacy of international arbitration as an adjudicative mechanism.
E. Protection of Fundamental Rights
The Checklist (¶ 129) highlights fundamental rights that are particularly at risk when AI systems are deployed, including data protection, non-discrimination, privacy, access to justice, and the right to a fair trial.
Arbitral tribunals must safeguard data protection, confidentiality, and equality of arms while mitigating biases inherent in AI systems used for evidence processing and decision support. To balance innovation with due-process guarantees, institutions can implement bias audits, ensure fair access to AI tools, and employ supervised technology sandboxes for controlled testing.
F. Effective Remedies and Contestability
The Checklist (¶ 130) requires accessible remedies enabling individuals to challenge the adoption, design, or use of AI systems in adjudication. Although arbitration offers narrower remedial pathways than domestic courts, analogous safeguards can be implemented.
Procedurally, parties may object to AI use via procedural motions, requesting disclosure, opposing specific tools, or seeking adjustments to preserve fairness. Substantively, if AI compromises due process, this may constitute grounds for annulment under institutional rules (e.g., ICSID Arbitration Rule 71) or non-enforcement under the New York Convention.
Parties must understand how AI influenced tribunal reasoning. These safeguards ensure that AI use in arbitration remains aligned with rule-of-law principles.
IV. Conclusion
The Venice Commission’s Rule of Law Checklist provides a principled and coherent framework for evaluating AI’s role in adjudication. Applying its standards to international arbitration highlights the need for legality, transparency, human-centered reasoning, rights protection, and effective remedies. As arbitral institutions and practitioners increasingly engage with AI systems and tools, adopting these principles can preserve the legitimacy, fairness, and enforceability of arbitral awards.
While arbitration enjoys flexibility unmatched by state courts, a fundamental comparative advantage, its legitimacy also depends on adherence to due process norms. The Checklist thus offers a valuable blueprint for ensuring that AI enhances, rather than undermines, the rule of law in the international arbitral context.
*Qerim Qerimi is Professor of International Law at the University of Prishtina and former Rector of the University. He serves as a member of the Venice Commission of the Council of Europe and as Chair of its Scientific Council. His research and publications focus on international law, international dispute settlement, human rights, rule of law, and international governance.