Sherzodbek Omonov & Saidrakhim Ibadullaev 1
AI is no longer viewed as a homogeneous and unified technology. It represents an ecosystem capable of processing content, generating recommendations and predictions, analyzing data, and identifying patterns. This includes machine learning tools that replicate certain aspects of human thinking, as well as natural language processing systems capable of interpreting and generating written and spoken language with surprising accuracy. These tools are already changing what practitioners expect from arbitration and how far the process can be improved, as illustrated by the American Arbitration Association’s AI arbitrator, currently being piloted in selected construction cases. Recent surveys already confirm that most practitioners believe AI will become a standard tool for legal research, document analysis, and case management within the next five years. This means not only a technological upgrade, but also rethinking the basic competencies expected of party representatives and arbitrators, as well as regulatory frameworks governing the use of AI.
Historically, arbitration has influenced the practice of traditional courts, sometimes referred to as the “arbitralization of courts.” This has resulted in a practice where courts act rather as adopters, while arbitration assumes the role of creator, serving as a laboratory for experimentation. The implementation of electronic signatures demonstrates this dynamic, having been the subject of skepticism in judicial practice mainly over so-called “non-documentary evidence,” while international arbitration bodies began accepting electronic notifications and electronic signatures to support business.Whether it be virtual hearings or issues surrounding evidence in commercial disputes, arbitration has formed sustainable practices and standards of conduct, often influencing the shaping of public law practices. It is against this backdrop that the question arises whether arbitration will once again become a testing ground – this time for the regulation of AI – or whether, in developing jurisdictions like Uzbekistan, the state courts will be the first to step into that role. Ultimately, the question is which institutional environment will be capable not only of implementing AI but also of developing principles for its responsible and legitimate use.
Uzbekistan, second only to its Central Asian neighbor in the government’s AI readiness, is already experimenting with AI tools in its judicial system. The “Digital Court” concept includes using AI to predict the outcomes of court proceedings and assess related litigation costs, as well as generating real-time text transcripts of court hearings. Initially, this will be tested in economic, civil, and administrative disputes – areas that overlap with many of the disputes typically addressed in commercial and investment arbitration. Still, this overlap may point to something more than a thematic parallel. Court proceedings generate the volume and public record that AI models need, while arbitration, given its confidential nature, does not. In turn, there is little to train on, and court-based experimentation may be the closest thing to a proving ground that AI tools in this space are likely to get.
In parallel, Uzbekistan is reinforcing its AI policy through legislative acts, as evidenced by the AI law changes. While the law sets the general direction of state policy, it does not address responsibility for algorithmic errors or procedural guarantees for the parties involved. This, in turn, does not resolve the procedural and institutional issues that arise when using AI in judicial activities. A situation is emerging in which courts are effectively the first to encounter AI in practice, even when it is used only as an auxiliary tool, while the rules for its application continue to be developed in a fragmented, retrospective manner.
Paradoxically, legislative changes in the field of arbitration place Uzbekistan in a rare and strategically important position, as the country simultaneously strives for deep integration into international arbitration practice and forms its key arbitration institutions, embodied by TIAC, in conditions of relative institutional “youth.” It is this combination that creates a space of opportunity that is less accessible to more mature arbitration centers bound by established procedures, professional habits, and technological “legacy.” For Uzbek arbitration, this creates not only a mechanism for economic and regulatory integration into the global community but also a platform for the proactive formation of policy in the field of technology management, primarily the adoption of AI.
Arbitration offers different risk management models to be tested without the danger of setting harmful precedents for the entire judicial system. The most effective solutions in Uzbekistan’s arbitration practice, such as protocols for verifying AI-generated evidence, could eventually serve as reference points for the broader international arbitration community. The inevitable moment of truth will come at the stage of recognition and enforcement of arbitral awards. This is where the state courts of Uzbekistan willtruly confront the results of AI’s application in arbitration. If the tribunal used AI to prepare a draft decision or analyze the facts, the losing party may try to challenge such a decision.
The grounds for challenge are, however, not uniform. Where AI is employed merely to organize or summarize factual submissions – material already on the record – the objection is arguably weaker, particularly if the arbitrator independently verifies and adopts the output. The more serious concern arises where AI generates the reasoning itself, as this touches the core of the adjudicative function that the parties entrusted to the arbitrator, not to a machine. In such cases, the challenge may rest not only on a violation of “public order” (Article 52 of the 2021 Law on International Commercial Arbitration) but also, and perhaps more directly, on the ground that the composition of the arbitral tribunal or the arbitral procedure did not conform to the parties’ agreement (Article 50(2)(1)), since the parties consented to resolution by named, human decision-makers.
Further, according to the Economic Procedural Code of Uzbekistan and the 2023 clarifications of the Supreme Court Plenum, Uzbek courts must adhere to the New York Convention and limit interference in arbitration. However, the use of AI introduces new variables: does “algorithmic assistance” comply with the “human-in-the-loop” principle enshrined in the 2026 AI legal landscape, which requires significant decisions where rights are impacted? The Supreme Court will face the need to interpret ethical norms of AI in the context of the arbitration process. In cases involving AI, courts will have to determine where AI’s “technical assistance” ends and the impermissible delegation of arbitral functions to a machine begins. This dialogue between flexible tribunals and conservative courts will be a key factor in building trust in Uzbekistan as a modern seat of arbitration.
The 2026 AI Amendments Law establishes a framework for all market participants. At the heart of this system is the human being and their dignity. Key requirements include the following:
- Prohibition of harm: AI-generated information resources and AI-powered information systems must not cause harm to a person’s life, health, freedom, honor, or dignity, nor violate any other inalienable rights—a norm that, while broadly framed, carries concrete legal force: any AI output that infringes these rights exposes its operator to liability.
- Human-in-the-loop: The law further reinforces the requirement discussed above—prohibiting reliance solely on AI-generated conclusions when making legally significant decisions affecting individual rights and freedoms.
- Protection of personal data: Strict administrative penalties are introduced for the unlawful processing of personal data through AI systems, which is especially relevant for arbitration cases involving international corporations.
For practitioners, this means a need to audit the tools used. Take, for example, an arbitrator who uses AI to assess witness credibility or estimate damages. They would need to be ready to explain why the tool is reliable and show that it carries no hidden biases. But there is a more fundamental question beneath all of this: whether using AI at all, especially without telling the parties, is even consistent with the arbitrator’s role as the parties understood it when they agreed to have their dispute resolved by a specific human being.
The arbitrator’s immunity, enshrined in the 2021 Law on International Commercial Arbitration, protects them from civil liability for actions taken in good faith, but does not exempt them from the obligation to comply with fundamental ethical norms of AI. One might reasonably ask whether using AI without disclosing it can be considered good faith at all. Of course, keeping it undisclosed could undermine the very foundation of the immunity, and it may well create a duty to be transparent with the parties, as a natural extension of the arbitrator’s obligation to remain independent and impartial.
The question of whether Uzbekistan will become a pilot jurisdiction for AI governance in international arbitration may already be answering itself. The unique combination of a “young” legal framework and the digital nature of TIAC creates optimal conditions for the first chapters of real-world AI practice to unfold precisely in arbitration halls. While state courts will understandably proceed through incremental digitalization of established processes, international arbitration in Uzbekistan may be called upon to resolve disputes at the frontier of technological change. The question that remains is whether arbitral practice will evolve flexibly enough to supply coherent standards–or whether it will expose the limits of private adjudication in regulating AI-driven conduct.
- Sherzodbek Omonov, LLM (2026, Harvard)
Saidrakhim Ibadullaev, Associate at Dentons in Tashkent ↩︎