{"id":389391,"date":"2026-07-04T15:30:22","date_gmt":"2026-07-04T10:00:22","guid":{"rendered":"https:\/\/www.scconline.com\/blog\/?p=389391"},"modified":"2026-07-04T15:30:07","modified_gmt":"2026-07-04T10:00:07","slug":"ai-in-arbitration-iac-eaw26-ai-governance-ethics-accountability","status":"publish","type":"post","link":"https:\/\/www.scconline.com\/blog\/post\/2026\/07\/04\/ai-in-arbitration-iac-eaw26-ai-governance-ethics-accountability\/","title":{"rendered":"\u201cTrust, but verify\u201d: Experts discusses AI governance, ethics and accountability at IAC Eurasia Arbitration Week 2026"},"content":{"rendered":"
\n

At the International Arbitration Centre (IAC) Eurasia Arbitration Week 2026 (EAW26), held in Astana, Kazakhstan, a panel titled “AI in Arbitration: Appropriate Use, Limits & Ethics”<\/span> brought together practitioners from arbitral institutions, legal technology companies and private practice to discuss the use of artificial intelligence in arbitration. Moderated by Jue Li<\/span>, Chief Executive Officer of LegalGo Intelligence, the panel featured<\/p>\n

    \n
  1. \n

    Alexandre Vagenheim<\/span>, Vice President, Institutional & Legal Affairs, Jus Mundi;<\/p>\n<\/li>\n

  2. \n

    Ziping Wei<\/span>, Director of Supervision and Coordination Division and Secretary General of the South China Sub-commission, CIETAC;<\/p>\n<\/li>\n

  3. \n

    Filip Nordlund<\/span>, Legal Engineering Manager and Practice Lead for International Arbitration, Legora;<\/p>\n<\/li>\n

  4. \n

    Dr. Kabir Duggal<\/span>, Lecturer at Columbia Law School and Partner at Akin Gump Strauss Hauer & Feld LLP; and<\/p>\n<\/li>\n

  5. \n

    Christopher Campbell-Holt OBE<\/span>, Registrar and Chief Executive of the AIFC Court and the IAC.<\/p>\n<\/li>\n<\/ol>\n

    The discussion examined the growing role of artificial intelligence across the arbitral process, beginning with its practical applications in legal practice and institutional administration before turning to questions of transparency, confidentiality and decision-making. The panellists also explored issues relating to professional responsibility, AI-generated errors, regulatory approaches across jurisdictions and the implications of AI for procedural fairness and access to justice.<\/p>\n

    AI adoption in Arbitration practice<\/p>\n

    Opening the discussion, Jue Li<\/span> reflected on the growing use of artificial intelligence and observed that many legal professionals are already incorporating AI into their daily work. Against that backdrop, Jue<\/span> invited the panellists to examine the relationship between AI and arbitration, posing questions on how AI is being used in arbitral practice, the opportunities it presents and the limits on its use. As the starting point for the discussion, Jue<\/span> invited the panellists to share their observations on how AI technology is being applied in their own work and practice.<\/p>\n

    \"\"<\/p>\n

    Responding to the opening question, Alexandre Vagenheim<\/span> began by asking what AI actually does in arbitration. He observed that there are several ways of answering that question. While one approach is to identify the individual tasks AI performs, such as legal research, case management and document summarisation, Alexandre<\/span> remarked that this is a limited way of understanding its role.<\/p>\n

    According to Alexandre<\/span>, the more significant development is the shift from individual tasks to broader workflows. He observed that AI is increasingly being embedded throughout arbitration processes, including case management, document organisation and other stages of the arbitral lifecycle. Referring to developments across the legal sector, Alexandre<\/span> remarked that both general-purpose and domain-specific legal AI are becoming integrated into arbitration workflows, reflecting a broader transformation in how legal work is carried out.<\/p>\n

    Alexandre<\/span> further observed that there is a third way of looking at AI, namely, how it is changing the mode of operation within the arbitration process. He remarked that, traditionally, legal reasoning has been at the core of a lawyer’s role. However, in his experience, AI should not be viewed merely as a tool for executing tasks but as an analytical platform that supports legal reasoning. Referring to his own use of AI, Alexandre<\/span> observed that it can help practitioners move beyond their initial analysis by serving as a collaborative tool for brainstorming and analytical thinking rather than simply automating legal work.<\/p>\n

    \"\"<\/p>\n

    Concluding his remarks, Alexandre<\/span> observed that the tools currently being developed increasingly reflect this collaborative approach to the use of AI in arbitration.<\/p>\n

    Responding to the opening question, Ziping Wei<\/span> observed that AI is transforming a range of industries, including arbitration. She remarked that, while discussions previously centred on whether AI would replace legal professionals, the focus has now shifted to how AI can be effectively integrated into and transform daily work.<\/p>\n

    Turning to CIETAC’s experience, Ziping<\/span> observed that, in response to the growing demand for digitalisation, CIETAC has developed a digital ecosystem over the past several years. She noted that the institution introduced an online case-filing system in 2019, a video hearing system in 2020, an online document exchange platform in 2023, and an integrated smart arbitration application in 2025. According to Ziping<\/span>, these systems collectively form CIETAC’s online dispute resolution (ODR) platform, enabling paperless administration of arbitral proceedings from filing and document exchange through hearings to the signing and issuance of awards.<\/p>\n

    Referring to the platform’s use, Ziping<\/span> observed that, in the previous year, 3,386 cases were filed online, representing approximately 60% of the institution’s annual caseload. She further noted that virtual or hybrid hearings were conducted for parties from 52 countries and regions, while 445 cases were handled entirely through the ODR platform.<\/p>\n

    Turning to the practical applications of AI, Ziping<\/span> identified translation, simultaneous interpretation and real-time transcription as areas where AI has improved efficiency. She observed that, with the assistance of AI products available in the Chinese market, CIETAC has been developing interpretation and transcription services to assist parties and tribunals during hearings conducted in English and Chinese. According to Ziping<\/span>, the accuracy of these services improves following training.<\/p>\n

    \"\"<\/p>\n

    Ziping<\/span> further observed that AI has made the selection of arbitrators more efficient and transparent. She noted that CIETAC maintains a database containing information on arbitrators, including their nationality, languages, areas of expertise and dynamic performance indicators, such as case-handling experience, current caseload and conflict-of-interest profiles. According to Ziping<\/span>, AI assists in matching party requirements with suitable arbitrators, thereby facilitating tribunal appointments.<\/p>\n

    Turning to future developments, Ziping<\/span> observed that CIETAC is working on AI tools capable of extracting arguments from parties’ submissions, generating procedural and factual sections of awards, identifying disputed and undisputed issues, and assisting tribunals with the preliminary stages of award drafting. She remarked that, to support these functions, CIETAC is digitising case files decided before 2024 for AI model training, intending to expand the capabilities of its ODR platform.<\/p>\n

    Responding next, Filip Nordlund<\/span> observed that AI is currently being embedded throughout the arbitration world, including by arbitral institutions, courts, barristers, arbitrators, parties and both global and large law firms, with different users employing AI at different levels and in different circumstances.<\/p>\n

    \"\"<\/p>\n

    Referring to Alexandre’s<\/span> earlier remarks on workflows, Filip<\/span> observed that one question going forward is whether practices may begin to change depending on the AI tools being used. He remarked that this could influence areas such as legal research, drafting and the way issues are addressed in submissions, depending on the tools an arbitrator may use to summarise those submissions.<\/p>\n

    According to Filip<\/span>, although this may sound like science fiction or something still far away, he believes it is already becoming a reality. He observed that some barristers and arbitrators are using AI tools to summarise submissions before considering a matter. In that context, Filip<\/span> remarked that practitioners should think about the tools being used and the effect they might have.<\/p>\n

    Responding to the question, Dr. Kabir Duggal<\/span> observed that, from the perspectives of both academia and legal practice, AI is already influencing arbitration in significant ways. Referring to tools such as Jus Mundi, Legora and Harvey, he remarked that “these programs can do what very smart associates do very quickly”, including reviewing thousands of pages of documents and producing summaries within minutes.<\/p>\n

    At the same time, Dr. Duggal<\/span> cautioned that AI-generated outputs should be verified. Referring to the expression “trust but verify”, he observed that, despite the capabilities of these tools, users should continue to validate the results they produce.<\/p>\n

    Turning to legal practice, Dr. Duggal<\/span> observed that clients are increasingly asking law firms how they intend to use AI. According to Dr. Duggal<\/span>, companies are also viewing AI as a cost-saving measure, indicating that AI is becoming an established part of legal practice.<\/p>\n

    Turning to arbitrators, Dr. Duggal<\/span> observed that there appears to be an emerging, informal understanding of acceptable and unacceptable uses of AI. He remarked that delegating arbitral decision-making to AI is generally regarded as unacceptable, while noting that the issue becomes more nuanced when considering the role of AI in assisting arbitrators.<\/p>\n

    Reflecting on academia, Dr. Duggal<\/span> observed that universities have adopted a cautious approach towards AI. Referring to his experience at Columbia University, he remarked that institutions have, in some instances, restricted the use of AI because of the speed with which it has developed and its ability to perform academic tasks. According to Dr. Duggal<\/span>, this approach is likely to evolve over time.<\/p>\n

    \"\"<\/p>\n

    Illustrating his broader point through an analogy, Dr. Duggal<\/span> observed that, in the 1870s, London’s concern over horse manure disappeared with the arrival of the motor car. He remarked that AI presents a similar moment of technological change, observing:<\/p>\n

    “AI is not going to take us back to the way we did things. The only way is going forward, it’s just how we go forward.”<\/p>\n

    Responding to the moderator’s opening question, Christopher<\/span> observed, with a touch of humour, that he was pleased to see a full audience despite hearing references to the previous night’s celebrations, remarking that he was curious why he had not been invited.<\/p>\n

    Turning to the role of AI, Christopher<\/span> observed that the AIFC Court and the IAC seek to work with the wider legal and technology communities to harness modern digital technologies in support of dispute resolution. According to Christopher<\/span>, the objective is not to replace judges or arbitrators but to identify the most appropriate ways in which technology can assist judicial and arbitral processes.<\/p>\n

    Reflecting on the early years of the AIFC Court and the IAC, Christopher<\/span> recalled that, shortly after he joined the institution around ten years earlier, someone had suggested that technology would eventually eliminate the need for judges and arbitrators. He remarked that, even then, he did not share that view and observed that discussions about the role of technology in dispute resolution have been ongoing for many years.<\/p>\n

    According to Christopher<\/span>, the real challenge is not simply identifying the best available technology but determining what functions technology should perform and how it can make a genuine contribution to improving judicial and arbitral processes.<\/p>\n

    Turning to practical applications, Christopher<\/span> observed that the AIFC Court and the IAC have been working to advance the use of AI and digital technologies in Central Asia. Referring to the conference itself, he noted the live transcription system providing simultaneous English and Russian transcription, which he stated was approximately 90 to 95 per cent accurate.<\/p>\n

    According to Christopher<\/span>, the system represented a world first for instant bilingual transcription in English and Russian and was developed with the assistance of digital technology experts in Kazakhstan.<\/p>\n

    \"\"<\/p>\n

    Christopher<\/span> further observed that the AIFC Court and the IAC intend to deploy this technology in case management conferences and hearings, whether conducted virtually or in person.<\/p>\n

    Turning to broader digital initiatives, Christopher<\/span> observed that different jurisdictions use terms such as “internet courts”, “e-justice” and “smart courts”, but remarked that these initiatives ultimately pursue the same objective of using technology to support dispute resolution. According to Christopher<\/span>, technology should be used to improve efficiency, reduce delays and costs, and enhance accuracy, while recognising that its use also raises important considerations.<\/p>\n

    Finally, Christopher<\/span> observed that legal technology providers are increasingly developing early neutral evaluation and case prediction tools. He remarked that such technologies are already assisting parties, particularly in facilitating early settlements and reducing unnecessary burdens on courts and arbitration centres. According to Christopher<\/span>, if technology is used effectively across these areas, it has the potential to make a significant contribution to the administration of justice.<\/p>\n

    Transparency and disclosure in AI-Assisted Arbitration<\/p>\n

    Turning to a more specific issue, Jue<\/span> observed that AI tools are already being used in arbitration practice and referred to the live AI-powered translation demonstrated during the session as one example of their application. Against that backdrop, Jue<\/span> turned the discussion to the issue of transparency, asking where the line should be drawn in relation to the disclosure of AI use in arbitral proceedings. Jue<\/span> further invited Ziping<\/span> and Christopher<\/span> to comment on whether CIETAC and the AIFC had considered introducing rules requiring such disclosure.<\/p>\n

    Responding to the moderator’s question, Ziping<\/span> observed that transparency forms a core procedural foundation for fairness and enforceability in arbitration. According to Ziping<\/span>, efficiency and impartiality constitute the two fundamental pillars of arbitration, with efficiency serving as the driving force and impartiality providing the foundation. She remarked that, although AI offers significant advantages in terms of efficiency and convenience, it must not undermine the fundamental principles of fairness, impartiality and due process.<\/p>\n

    Turning to the disclosure of AI use, Ziping<\/span> observed that disclosure should be required where the use of AI materially affects procedural fairness or creates a risk that an arbitral award could be set aside or refused enforcement. At the same time, she remarked that AI is currently used in arbitration in a wide variety of ways and that the extent of disclosure should therefore depend on the particular circumstances.<\/p>\n

    Drawing that distinction, Ziping<\/span> observed that where AI is used only for administrative functions, such as electronic service or remote hearings, disclosure requirements may appropriately be more limited. By contrast, she remarked that comprehensive disclosure should be made where AI is involved in core adjudicatory functions, including evidence evaluation, allocation of liability, assessment of quantum, legal reasoning or other decision-making processes. According to Ziping<\/span>, such disclosure should extend to the type of AI used, the scope of input data, the generation logic and the basis of its conclusions so as to preserve transparent reasoning and procedural fairness.<\/p>\n

    Turning to the risks associated with AI, Ziping<\/span> referred to what she described as the “black box” problem, observing that where the reasoning process is inaccessible, parties may be unable to effectively challenge the AI’s analysis of evidence or legal issues, thereby affecting due process. She also referred to AI hallucinations, observing that AI may generate non-existent legal authorities, case details or factual conclusions. According to Ziping<\/span>, if such errors are not identified through proper verification, they may lead to procedural irregularities, the setting aside or non-enforcement of arbitral awards, and may undermine the finality and legitimacy of arbitration.<\/p>\n

    Referring to an arbitral award that she understood had been affected by AI hallucinations, Ziping<\/span> concluded on a lighter note that she had located the case through Google and hoped “it’s a real, not a made-up one.”<\/p>\n

    Following Ziping’s<\/span> remarks, Jue<\/span> invited Christopher<\/span> to respond from the perspective of the parties and to address whether arbitral tribunals should require parties to disclose their use of AI during arbitral proceedings.<\/p>\n

    Responding, Christopher<\/span> observed that, in his personal view, parties should disclose their use of AI in arbitral proceedings. According to Christopher<\/span>, arbitration is currently undergoing a period of significant change, and the overriding objective should remain the administration of justice through the appropriate use of digital technologies. He remarked that transparency, equality and fairness should remain central to dispute resolution and that, given the pace of technological change, the interests of justice are best served by being “as transparent as humanly possible.”<\/p>\n

    Turning to the broader regulatory landscape, Christopher<\/span> observed that there is not yet a global consensus on the appropriate use of AI in dispute resolution. While noting that numerous codes of practice, guidance documents and rules are continuing to emerge, he remarked that greater collective agreement is needed on both the appropriate and inappropriate uses of AI in arbitral proceedings.<\/p>\n

    Referring to developments at the AIFC Court and the IAC, Christopher<\/span> observed that the institutions are engaged in preliminary discussions, both internally and with stakeholders, on developing guidance for the appropriate use of AI and digital technologies in dispute resolution proceedings. According to Christopher<\/span>, the intention is to issue practice directions for both the Court and the Arbitration Centre and, where appropriate, incorporate those requirements into their procedural rules.<\/p>\n

    Christopher<\/span> further observed that existing AI guidelines often set out only broad principles without providing sufficient practical direction. According to Christopher<\/span>, the guidance being considered seeks to define more clearly the standards expected in the use of AI while remaining consistent with the overriding objective of administering justice.<\/p>\n

    Turning to the matters under consideration, Christopher<\/span> observed that the proposed guidance aims to ensure that AI does not disadvantage users with limited access to digital tools or digital skills and that procedural fairness remains protected. He remarked that parties should be able to understand, respond to and test AI-assisted material relied upon during proceedings.<\/p>\n

    According to Christopher<\/span>, one proposal under consideration is that parties should declare, at the earliest possible opportunity, whether they have used or intend to use AI-generated content during arbitral proceedings. He further observed that any concerns regarding the use of AI should, where possible, be resolved at the case management conference stage and, where agreement cannot be reached, be referred to the arbitral tribunal for determination.<\/p>\n

    Christopher<\/span> also remarked that parties should not wait until shortly before a hearing to disclose their intended use of AI-generated content, as this could result in adjournment requests and unnecessary delays. He further observed that where AI is used to assist with legal research, interpretation of facts or the application of law to facts, prompt written disclosure should be made to all relevant parties.<\/p>\n

    Illustrating the level of disclosure being considered, Christopher<\/span> observed that such disclosure may include the name and version of the AI system, the relevant functional settings, a description of the AI-generated or AI-assisted content, the complete prompt history, including any templates or contextual information provided to the system, and the resulting outputs. According to Christopher<\/span>, the objective is to encourage appropriate disclosure throughout the proceedings so that all participants understand how AI has been used and to promote a fairer and more transparent process.<\/p>\n

    Turning first to the previous discussion on disclosure, Alexandre<\/span> observed that there may be areas where consensus exists and others where views continue to differ. He remarked that the question of disclosure also depends on the role a person performs in an arbitration, observing that counsel and arbitrators have different responsibilities.<\/p>\n

    From the perspective of counsel, Alexandre<\/span> observed that lawyers should have every available tool at their disposal to represent their clients, including AI tools. He remarked that he does not believe counsel should be required to disclose every use of AI, observing that there has traditionally been no obligation to disclose the use of other tools in legal practice. At the same time, Alexandre<\/span> acknowledged that issues affecting the integrity of arbitral proceedings may require discussion and observed that current regulatory initiatives are examining how those questions should be addressed within the arbitral process.<\/p>\n

    Turning to arbitrators, Alexandre<\/span> observed that the position is different because arbitrators exercise a decision-making function. He remarked that this is where many of the current discussions are centred. Referring to the IBA AI Guidelines, on which he stated he was working as part of a group of 30 members chaired by the LCIA President, Alexandre<\/span> observed that the objective is to develop guidance for specific use cases concerning disclosure at different stages of the arbitral workflow. He also referred to an initiative of the Silicon Valley Arbitration & Mediation Center (SVAMC) to develop an AI protocol for arbitration, observing that it likewise focuses on issues relating to AI and decision-making.<\/p>\n

    Turning to confidentiality, Alexandre<\/span> observed that the primary responsibility for protecting confidential information rests with lawyers rather than legal technology providers. He remarked that lawyers owe a duty of confidentiality to their clients and distinguished that duty from issues relating to privacy settings and technical security measures.<\/p>\n

    From the perspective of a legal technology provider, Alexandre<\/span> observed that AI tools should enable legal practitioners to comply with their confidentiality obligations. Referring to the tools developed by Jus Mundi, he remarked that they are designed to comply with applicable regulatory requirements, including the EU AI Act, which treats the administration of justice as a high-risk use of AI. According to Alexandre<\/span>, legal technology providers should implement appropriate security, privacy and encryption measures, adding that no provider is likely to succeed unless it adequately addresses confidentiality, security and privacy requirements.<\/p>\n

    Following the discussion on confidentiality, Jue<\/span> turned to the use of AI by arbitrators and invited the panellists to reflect on the appropriate limits of AI in arbitral decision-making, including which uses of AI are generally regarded as acceptable and where human judgment should remain indispensable.<\/p>\n

    Responding first, Filip<\/span> observed that legal technology providers adopt different technical approaches to protecting confidential information. Referring to the security measures available in AI systems, he noted that different data storage models and security arrangements may be employed depending on the system being used. According to Filip<\/span>, confidentiality also requires users to understand how an AI system operates and the manner in which information is handled within that system.<\/p>\n

    Human Decision-Making and the Limits of AI<\/p>\n

    Responding, Dr. Duggal<\/span> observed that generative AI remains a relatively recent development, having emerged only in late 2022, but has already transformed legal practice. Turning to arbitral decision-making, he remarked that one principle on which there appears to be broad consensus within the legal community is that decision-making should not be delegated to AI. According to Dr. Duggal<\/span>, this principle is reflected in various guidance documents, including those developed by the Silicon Valley Arbitration & Mediation Center (SVAMC).<\/p>\n

    At the same time, Dr. Duggal<\/span> observed that developments continue in this area. Referring to the American Arbitration Association, he noted that it has introduced an AI arbitrator for certain low-value construction disputes, with a human reviewer remaining involved in the process. He remarked that the legality of such an approach remains to be tested and observed that, where parties freely agree to such mechanisms, the issue may ultimately require further legal consideration.<\/p>\n

    Turning to AI-assisted document review, Dr. Duggal<\/span> observed that differing views exist within the arbitration community. He remarked that some consider AI-assisted summarisation and chronology preparation to be comparable to work undertaken by tribunal secretaries, while others maintain that the selection and evaluation of relevant material should remain the responsibility of the arbitrator whose judgment the parties have chosen. According to Dr. Duggal<\/span>, the arbitration community has yet to reach a consensus on this issue.<\/p>\n

    Concluding his remarks, Dr. Duggal<\/span> cautioned practitioners to verify AI-generated outputs before relying on them. Referring to Benjamin Franklin’s advice to “trust, but verify”, he observed that AI can sometimes generate inaccurate authorities or information with apparent confidence, making human verification essential.<\/p>\n

    Building on the discussion, Christopher<\/span> agreed that arbitral decision-making should remain a human function. Referring to discussions among judges, arbitrators and academics, he observed that there appeared to be broad agreement on the importance of human judgment and emotional capacity in dispute resolution. According to Christopher<\/span>, parties who appoint an arbitrator expect not merely the output of a machine but the independent judgment and decision of a human decision-maker.<\/p>\n

    Turning to document review, Christopher<\/span> observed that the use of AI to assist in managing large volumes of documents has become increasingly common in major disputes. Referring to a matter involving approximately 15 million documents, he remarked that the use of AI had enabled the legal team to reduce the volume of material to between five and six million documents, which he described as a constructive outcome that did not appear to hamper the dispute resolution process.<\/p>\n

    Returning to the broader issue of procedural fairness, Christopher<\/span> observed that the challenge lies in preserving the integrity of arbitral proceedings while making appropriate use of AI. He remarked that greater guidance is needed on ensuring the accuracy of material presented to decision-makers, particularly where proceedings involve extensive documentary evidence and constraints of time and cost.<\/p>\n

    According to Christopher<\/span>, professional duties relating to the independent verification, accuracy and authenticity of documents submitted to arbitral tribunals should receive greater attention, particularly where AI has been used. He also reiterated his support for disclosure requirements concerning the use of AI in arbitral proceedings.<\/p>\n

    Christopher<\/span> further observed that parties and their representatives should remain responsible for any uncorrected errors or inaccuracies in material submitted with the assistance of AI. He also suggested that parties should reach a clearer understanding at the outset of arbitral proceedings regarding the standards they are prepared to accept in relation to the accuracy and legitimacy of AI-assisted evidence, so that all participants proceed on an equal footing.<\/p>\n

    AI hallucinations, accountability and professional responsibility<\/p>\n

    Turning to the issue of inaccurate AI-generated outputs, Jue<\/span> asked who should bear responsibility where AI produces incorrect citations, statements or other information.<\/p>\n

    Responding, Filip<\/span> observed that the issue involves two aspects, beginning with the question of responsibility. According to Filip<\/span>, lawyers are responsible to their clients and remain accountable for the submissions they make. He remarked that where a lawyer cites a non-existent case or submits incorrect information generated by AI, responsibility rests with the lawyer.<\/p>\n

    From the perspective of a legal technology provider, Filip<\/span> observed that the responsibility of AI companies is to ensure the highest possible level of accuracy for users of their tools. At the same time, he remarked that users should continue to “trust but verify”, observing that this message is consistently emphasised during user training.<\/p>\n

    Concluding his remarks, Filip<\/span> observed that AI is a probabilistic rather than a deterministic tool. Explaining the distinction, he remarked that AI generates what it considers the most likely answer, and not necessarily the most correct one. According to Filip<\/span>, although the most likely and the correct answer will often coincide, users should remain aware of this inherent limitation of AI systems.<\/p>\n

    Responding, Alexandre<\/span> observed that, despite growing awareness of AI hallucinations, many lawyers continue to fail to verify AI-generated outputs. Referring to what he described as the first widely publicised hallucination case involving a New York lawyer in the Avianca<\/span> litigation, Alexandre<\/span> observed that the issue has since become much more widespread.<\/p>\n

    According to Alexandre<\/span>, there are now more than 1,500 reported instances worldwide in which fake AI-generated case citations have been submitted before courts. He also referred to an open-access database created by French lawyer Damien Charlotin<\/span>, which catalogues cases involving the submission of fabricated or incorrect AI-generated citations.<\/p>\n

    Turning to the nature of AI hallucinations, Alexandre<\/span> observed that they occur in varying degrees. While the most obvious examples involve entirely fabricated case citations, he remarked that more subtle forms also exist, including incorrect citations or reliance on authorities that are not appropriate for the legal proposition being advanced.<\/p>\n

    Reflecting on these developments, Alexandre<\/span> observed that the large number of reported cases demonstrates that lawyers continue to neglect the verification of AI-generated outputs, describing the situation as “really, really concerning”. At the same time, he remarked that these incidents also illustrate how persuasive and authoritative AI-generated content can appear.<\/p>\n

    Turning to the role of legal technology providers, Alexandre<\/span> observed that developers can continue improving the tools available to legal practitioners. Referring to Jus Mundi’s AI system, he remarked that it provides source-linked citations and achieves approximately 95 per cent accuracy in matching generated statements with their underlying sources. Nevertheless, Alexandre<\/span> observed that a margin of error remains and concluded that ensuring the accuracy of AI-generated material is a responsibility shared by the legal community.<\/p>\n

    Building on the discussion, Dr. Duggal<\/span> observed that the issue ultimately returns to first principles of professional ethics. Referring to lawyers practising in jurisdictions such as New York and England, he remarked that legal practitioners are under a professional duty not to mislead the court. According to Dr. Duggal<\/span>, failing to verify AI-generated authorities before relying on them constitutes a fundamental breach of that professional obligation.<\/p>\n

    Turning to recent developments, Dr. Duggal<\/span> observed that problems relating to AI hallucinations had initially been seen largely in smaller and mid-sized law firms, where lawyers often work under significant time pressures. However, he remarked that this was no longer the case. Referring to a recent bankruptcy proceeding, Dr. Duggal<\/span> observed that a major law firm had also encountered difficulties after relying on hallucinated case citations and had subsequently apologised.<\/p>\n

    Dr. Duggal<\/span> further referred to a recent decision of a court in Montreal, observing that an arbitral award had been annulled after the arbitrator relied on AI-generated, fabricated citations. According to Dr. Duggal<\/span>, the court held that the arbitrator’s conduct violated Canadian public policy. He remarked that, while previous discussions had focused on the obligations of counsel, similar concerns may also arise in relation to arbitrators.<\/p>\n

    Concluding his remarks, Dr. Duggal<\/span> reiterated the importance of verifying AI-generated outputs before relying on them, observing, “Trust, but verify.”<\/span><\/p>\n

    \"\"<\/p>\n

    Building on the previous example, Alexandre<\/span> observed that he had traced the identity of the arbitrator from the publicly available judgment and had reviewed a LinkedIn post in which the arbitrator explained the incident. According to Alexandre<\/span>, the arbitrator had sought to justify the use of the fabricated citation by stating that it had not been relevant to the legal reasoning and that the award therefore should not have been annulled. Alexandre<\/span> remarked that he found this explanation “quite shocking” and observed that, while the arbitrator was attempting to justify his conduct, the incident nevertheless underscored the seriousness of AI hallucinations and the need for legal practitioners to continue exercising diligence.<\/p>\n

    Regulating AI in Arbitration<\/p>\n

    Turning to the broader regulatory landscape, Jue<\/span> observed that the discussion had highlighted differing approaches across jurisdictions and invited the panellists to consider whether a uniform standard should be developed to regulate the use of AI in arbitral proceedings.<\/p>\n

    Responding, Ziping<\/span> observed that, despite the challenges discussed during the session, she did not believe a uniform standard regulating the use of AI in arbitration would be appropriate in the long term. Referring to the diverse ways in which AI is used across different jurisdictions and arbitral institutions, Ziping<\/span> remarked that it would be difficult to develop a single international standard in the near future.<\/p>\n

    According to Ziping<\/span>, a more prudent approach would be to continue monitoring developments in AI and to accumulate practical experience across arbitral institutions before introducing formal, uniform regulations.<\/p>\n

    Turning to CIETAC’s approach, Ziping<\/span> observed that the institution issued its AI Guidelines<\/span> in July 2025, adopting what she described as a principled, party-centred approach that seeks to promote procedural efficiency while safeguarding fairness, transparency and human control. She further observed that the guidelines were issued as a trial version so that they could be revised over time in response to developments in AI technology.<\/p>\n

    Ziping<\/span> observed that the guidelines are founded on three basic principles: party autonomy, the assistive role of AI in adjudication and good faith. Explaining these principles, she remarked that party autonomy preserves the parties’ ability to determine the use of AI in arbitral proceedings, subject to the applicable rules and the consent of the parties. According to Ziping<\/span>, the assistive role principle recognises that AI should support tribunals in making fair and informed decisions but should not replace the tribunal’s independent judgment. She further observed that the principle of good faith requires parties and tribunals to use AI honestly and fairly without seeking to obtain an unfair advantage.<\/p>\n

    Turning to practical guidance, Ziping<\/span> observed that CIETAC has also prepared separate recommended lists for arbitrators and parties to reduce the risks associated with the use of AI. According to Ziping<\/span>, these recommendations address matters such as data security, transparency and the verification of AI-generated outputs, with the objective of assisting parties and tribunals in using AI more safely during arbitral proceedings.<\/p>\n

    Concluding her remarks, Ziping<\/span> observed that CIETAC encourages the development and use of AI technologies that comply with legal and ethical standards.<\/p>\n

    Responding, Christopher<\/span> observed that he would like to see more examples of guidance to assist institutions and parties in reviewing their practices and adopting appropriate approaches to the use of AI. According to Christopher<\/span>, greater collaboration is needed to develop broader consensus as AI technology and legal practice continue to evolve.<\/p>\n

    Turning to areas where common ground may already exist, Christopher<\/span> observed that issues such as confidentiality, security, the use of AI in relation to evidence, witness statements, expert evidence, disclosure obligations and the consequences of non-compliance by parties and legal representatives are matters on which institutions can begin developing principled guidance. He remarked that these are issues on which meaningful examples and common approaches can already be developed.<\/p>\n

    Concluding his remarks, Christopher<\/span> referred to observations made during the opening session of the conference, noting that, while the legal community seeks universal standards, those standards must also take account of the different circumstances and environments in which institutions operate, particularly in rapidly developing jurisdictions such as Central Asia.<\/p>\n

    AI, equality of arms and access to justice<\/p>\n

    Turning to a final question, Jue<\/span> invited the panellists to consider whether an imbalance may arise where one party uses advanced AI tools during arbitral proceedings while the other does not, and asked whether such situations raise concerns of procedural fairness or require intervention by the arbitral tribunal.<\/p>\n

    Responding, Alexandre<\/span> observed that, from the perspective of legal technology providers, AI has the potential to level the playing field. Referring to the adoption of AI tools by smaller and medium-sized law firms, Alexandre<\/span> remarked that these firms were among the earliest users of such technologies and have been able to enhance their operational capacity. According to Alexandre<\/span>, AI enables those firms to process significantly larger volumes of documents and undertake larger arbitration matters than they might previously have been able to handle. He observed that, despite the cost of AI tools, their adoption presents an opportunity for smaller and regional firms to expand the scale of matters they are capable of undertaking.<\/p>\n

    Responding, Dr. Duggal<\/span> observed that this was one of the questions he had found most difficult. He remarked that, as officers of the law, legal professionals seek a system that is fair, just and capable of treating parties equally, irrespective of their resources.<\/p>\n

    At the same time, Dr. Duggal<\/span> observed that disparities already exist within legal practice, with some parties able to engage larger or more specialised law firms while others cannot. He remarked that he was uncertain where the line should be drawn in addressing similar disparities arising from the use of AI.<\/p>\n

    According to Dr. Duggal<\/span>, where the use of AI is clearly abusive, there may be a duty to intervene in order to preserve a level playing field. However, he also observed that arbitrators cannot be expected to resolve every inequality in society. Concluding his remarks, Dr. Duggal<\/span> acknowledged that he did not have a definitive answer to the question, joking that he wished he had “something sexier.”<\/p>\n

    Invited to comment, Christopher<\/span> observed that the primary objective of arbitration and dispute resolution is the administration of justice. According to Christopher<\/span>, while absolute equality is important, equal treatment remains a fundamental principle and is reflected in the overriding objective of both the Court and the Arbitration Centre.<\/p>\n

    Turning to AI, Christopher<\/span> observed that digital technology is the latest tool available to assist the administration of justice. He remarked that, if used appropriately, AI can contribute to the overall dispute resolution process, although the legal community must first reach greater agreement on what constitutes appropriate use.<\/p>\n

    According to Christopher<\/span>, the integrity and procedural fairness of arbitral proceedings must never be compromised through the use of AI. Referring to the earlier discussion on differences in the resources available to parties, Christopher<\/span> acknowledged that not all parties can afford to engage leading law firms. He observed that such inequalities can be addressed in some contexts, though not necessarily in high-value and complex disputes. He concluded by reiterating that the integrity of the system must not be compromised and that the ultimate objective remains the delivery of fair and just outcomes.<\/p>\n

    Responding, Filip Nordlund<\/span> observed that the use of AI has created what he described as a “two-tier system”, with a significant difference between those who have access to advanced AI tools and those who do not. According to Filip<\/span>, while AI was initially viewed primarily as a tool for improving efficiency through tasks such as summarisation and knowledge management, its capabilities have expanded considerably.<\/p>\n

    Reflecting on his own experience, Filip<\/span> observed that AI has changed not only the speed of legal work but also its quality. He remarked that AI enables practitioners to explore and develop multiple legal arguments simultaneously, conduct legal research alongside that process and evaluate different approaches within the limited time available to prepare a case. According to Filip<\/span>, these capabilities represent a significant change in legal practice.<\/p>\n

    Concluding his remarks, Filip<\/span> observed that the growing disparity in access to advanced AI tools is likely to have important implications for access to justice.<\/p>\n

    Bringing the session to a close, Jue<\/span> thanked the panellists for their contributions and observed that, despite the limited time available, the discussion had generated valuable insights.<\/p>\n

    Read more IAC Eurasia Arbitration Week 2026 coverage:<\/p>\n