scconline.com
GHAC Arbitration Week 2026: Experts examine the expanding role of AI in international and domestic arbitration
The third day of GHAC Arbitration Week 2026 turned to the growing role of technology and artificial intelligence in arbitral proceedings, examining how AI is already being used by counsel and arbitrators and the questions it raises around professional judgment, disclosure, confidentiality, arbitral decision-making and the training of younger lawyers.
The session, titled “Technology & AI in the Proceedings of International and Domestic Arbitration”, was moderated by Senior Advocate Devang Nanavati and featured:
Nakul Dewan, Senior Advocate and King’s Counsel;
Baiju Vasani, Barrister and Arbitrator, Twenty Essex;
Sameer Jain, Advocate and Indian Chair, ICC;
Rishab Gupta, Barrister, Indian Counsel and New York Lawyer; and
Aditya Singh, Partner, White & Case, Singapore.
AI is already embedded in arbitral practice
Opening the discussion, Devang Nanavati noted that technology was no longer merely assisting arbitration but was beginning to reshape how proceedings are conducted, including research, drafting, evidence and decision-making. He illustrated the growing capabilities of AI through an experiment using Claude to analyse judgments and predict the questions that judges might raise in a case. The exercise, he said, demonstrated how quickly AI could analyse judicial material and hearing content.
The discussion then moved to the extent to which AI is already being used in international arbitration. Rishab Gupta said that, in his experience, almost every disputes lawyer working in commercial dispute resolution, particularly international arbitration, was using AI in some form. He said he used it for summarising documents, working through large volumes of data, and editing and polishing drafts. In his view, a central issue was not simply whether AI was being used but how transparently it was being used.
Rishab Gupta referred to a recent arbitration seated in Geneva under the Swiss Rules in which the sole arbitrator told the parties at the first case management conference that he proposed to use AI tools instead of appointing a tribunal secretary. The arbitrator proposed recording the use of AI in the first procedural order and sought the parties’ consent. Rishab Gupta described this as an example of substantial transparency in the use of AI. He contrasted it with an arbitration award in Quebec which, according to an account he had read, was set aside after the arbitrator incorporated authorities and other material generated by AI that were not cited before the tribunal and included hallucinated citations.
Adoption of AI is being driven by efficiency and cost considerations
Aditya Singh said that he was yet to encounter a lawyer or arbitrator who did not use AI for some aspect of their work. Referring to the 2025 Queen Mary University of London and White & Case survey on international arbitration, he said that respondents had been asked about AI use across legal research, document review, correspondence, legal submissions and analysis of arguments and evidence. According to Singh, 91% of respondents expected to use AI for at least document review and legal research. He said the three principal reasons identified for using AI were saving time, reducing costs, and reducing human errors and inconsistencies.
“91% of the respondents said that they expected to use AI for at least document review and legal research.”
Aditya Singh also discussed the American Arbitration Association – International Centre for Dispute Resolution’s (AAA-ICDR) “AI Arbitrator”, which he described as an experiment presently limited to document-only construction arbitrations. He identified four features of the system: it is not intended for proceedings involving witness testimony because AI cannot presently assess witness credibility; both parties must consent; the system has been trained on more than 1,500 construction awards written by humans; and submissions uploaded by the parties are processed into summaries which are shared with counsel for review before the system generates a draft outcome. A human being then reviews the draft before the award is issued. Aditya Singh described it as a significant experiment in the potential use of AI in arbitral decision-making.
AI as assistance, not substitution
Baiju Vasani said that arbitrators and counsel should use AI and that there was “no shame” in doing so. He compared AI to other professional tools, including Google, and said the central question was how the tool was used and whether its use became an abuse.
“AI is a tool and not a replacement.”
For Baiju Vasani, the important distinction was between using AI with professional judgment and allowing AI to function by itself. Counsel were engaged for their experience, expertise and judgment, while arbitrators were appointed for their personal input. AI could therefore assist those functions, but should not become a means of abandoning them.
Nakul Dewan developed this distinction through an analogy with Google Maps. He said there was a difference between someone who knew the roads and used Google Maps to identify the quickest route and someone who simply followed the blue line without understanding the road network. The former retained the ability to recognise when the suggested route was wrong.
“When you use AI, if you substitute it for your own brain work, then you have a problem.”
Nakul Dewan described this risk in terms of “cognitive offloading”. Excessive reliance on AI could leave counsel without a sufficiently independent understanding of the case. In international arbitration, where cross-examination may take place within fixed hearing times and with live transcription, he said counsel needed to know the case file sufficiently well to respond immediately to unexpected developments. AI could therefore be used to expedite routine work, such as preparing chronologies, but the time saved should be used for deeper critical analysis.
Nakul Dewan later said that, despite having trained others in the use of AI, he had personally not yet used it extensively because of concerns about over-reliance and missing something in a case. He nevertheless acknowledged the need to adapt and said that younger lawyers should harness technology to save time while using that additional time for critical thinking.
“We need to harness it. We need to use that technology for the purposes of being able to save time and use that time for more critical thinking.”
Rishab Gupta similarly described AI as a productivity tool, rather than a replacement for lawyers. He stressed that human judgment and discretion remained essential.
AI in cross-examination: finding the “needle in the haystack”
The panel then considered how AI could assist in preparing and conducting cross-examination.
Rishab Gupta distinguished between proceedings before Indian courts, domestic arbitral tribunals and international tribunals. He noted that international arbitration generally operates within fixed hearing times, live transcription and limited opportunities to extend cross-examination. Counsel therefore had to know the case file thoroughly and remain ahead of the witness.
He said he personally used AI more heavily at the beginning of cases for understanding chronology and evidentiary problems and for drafting and editing. In cross-examination preparation, he identified two particularly useful applications: finding evidence hidden within very large datasets and generating ideas.
As an example, Rishab Gupta referred to a case involving approximately 20,000 pages of evidence and nearly 20,000 WhatsApp messages. While preparing the cross-examination of a particular witness, AI identified a sentence in a WhatsApp message indicating that the witness had met an individual on a particular day, contrary to the witness’s statement. Rishab Gupta said this was an example of AI finding a “needle in the haystack” that human reviewers had not found.
He also described using AI while preparing the cross-examination of a chemistry professor in a contaminated crude oil case. The tool generated possible lines of questioning which Rishab Gupta subsequently developed and assessed himself. He said AI could therefore be useful for ideation, much as counsel might discuss ideas with juniors or instructing lawyers.
Sameer Jain also acknowledged using AI to identify contradictions between witness evidence and documents. However, he cautioned that its output was not completely accurate. He said he initially tested the tool on cases that had already concluded, allowing him to assess whether the AI’s suggestions were heading in the correct direction. Sameer Jain emphasised that AI could not replace the human aspects of cross-examination, including observing how a witness responds, deciding whether a follow-up question was necessary and determining when to challenge credibility.
Baiju Vasani raised another dimension: witnesses themselves were beginning to use AI to anticipate questions. He observed that where witnesses used AI to prepare for cross-examination, there was a risk that they could become overly coached and consequently less useful to the tribunal or court.
Disclosure: different approaches for counsel and arbitrators
The question of whether AI use should be disclosed became a central part of the discussion.
In response to Devang Nanavati’s question about whether arbitral institutions or tribunals should require disclosure of the AI tools used by parties and counsel, Nakul Dewan said disclosure was important. He distinguished between research and drafting and highlighted the possibility that AI could affect the balance between parties with differently sized legal teams. At the same time, he considered AI capable of helping smaller teams narrow that gap.
Nakul Dewan also referred to the drafting of witness statements. In his view, a witness statement could not simply be generated by AI because the witness remained responsible for the factual account. Where AI was used, he said, the statement should be fact-checked by the witness. He referred to Singapore court guidelines concerning disclosure when AI is used for pleadings or witness statements.
Sameer Jain took a different position concerning counsel. He said it was too early to require general disclosure of the use of AI. In his view, pleadings, submissions, authorities and evidence had to stand on their own merits regardless of whether AI had assisted in their preparation. Referring to discussions in Singapore, he said that AI was still at an early stage and that Singapore had not, at that point, adopted a general regulatory approach to its use.
Sameer Jain’s position was distinguished by Aditya Singh, who said disclosure became imperative where an arbitrator was using AI. He drew a parallel with earlier controversy concerning tribunal secretaries and tribunal assistants, referring specifically to the Yukos v. Russia arbitration. He said arbitrators were selected for their expertise and personal decision-making, creating a stronger reason for transparency when technology was involved in that decision-making process.
Rishab Gupta agreed with this distinction. While he saw no difficulty in counsel using AI without disclosure in ordinary circumstances, he considered disclosure appropriate where arbitrators used it. He cited the Swiss arbitrator who had openly told the parties about his proposed AI use as an example of the transparency he considered appropriate.
Baiju Vasani similarly supported general procedural clarity but opposed requiring counsel to make granular disclosures every time AI was used. He said it was common for procedural orders to record that AI could be used by the tribunal and parties. In his view, the responsibility ultimately remained with the person who adopted the work product.
Confidentiality remains a separate and immediate concern
An audience member then raised a specific question on confidentiality: whether uploading client documents, pleadings and evidence to an AI platform could itself breach confidentiality obligations, particularly where arbitration proceedings were confidential.
Sameer Jain responded that the answer depended substantially on the architecture of the AI tool being used. He distinguished between public AI systems where uploaded information could be processed on external servers; legal AI systems that provide contractual or technical confidentiality safeguards; and models operating locally without an internet connection. He cautioned that using public platforms could result in a breach of confidentiality obligations.
Nakul Dewan added a fourth possibility: disclosure and consent. Where the parties and tribunal had agreed to the use of AI, he said, the confidentiality position could be addressed through that consent.
Aditya Singh gave an example from White & Case’s use of a private Legora system. He said the firm’s system provided controlled access and was designed so that uploaded information would not be used to train external models or appear in searches outside the system.
AI and Section 34 challenges
The panel then considered whether misuse of AI could become a basis for setting aside an arbitral award under Section 34 Arbitration and Conciliation Act, 1996, including on grounds such as patent illegality or perversity.
Baiju Vasani said AI did not, by itself, create a new ground for setting aside an award. The question remained whether the consequences of AI use fell within an existing ground of challenge. He gave the example of an AI hallucination and said the relevant inquiry would be whether the hallucination had a material effect on the decision.
He also distinguished adoption and delegation. If an arbitrator reviewed and adopted material produced with assistance from another source, that was different from simply delegating the adjudicatory function. In his view, the same conceptual distinction would apply where AI was used in preparing an award.
Devang Nanavati then shared an Indian example from litigation involving a Section 34 order. He said an order appeared to have used AI at least for its language, and a reference to a subsection of Section 34 was incorrectly rendered. The example was used to underline the possibility of AI-generated drafting errors even where the underlying judicial reasoning remained a human function.
Sameer Jain reinforced the distinction between assistance and abdication through an anecdote from a Global Arbitration Review quiz in Madrid involving the question of how many arbitrators were needed to change a light bulb. The winning answer was that none were needed because the tribunal secretary would do it. Sameer Jain said the underlying issue was whether the arbitrator was using assistance or abdicating responsibility, and whether the arbitrator had personally applied their mind to the reasoning and decision recorded in the award.
The training of young lawyers
Questions from the audience also focused on whether AI could deprive younger lawyers of the very work through which they traditionally develop legal instincts.
One audience member asked whether the removal of routine or “grunt” work by AI could undermine the training process, particularly because traditional research required junior lawyers to read a large number of authorities before finding the relevant case.
Nakul Dewan acknowledged the concern. He noted that legal research had itself evolved through successive technologies, from manual reporters and textbooks to online databases and search engines. The question, in his view, was not whether technology could be ignored, since it could not, but whether young lawyers would use it without developing their own understanding of legal propositions. He said junior lawyers should use the time saved by AI to read judgments more closely, develop propositions independently and improve their critical analysis.
The panel returned to this concern during the audience question-and-answer session. Sameer Jain said he did not expect AI to replace lawyers, but believed lawyers who refused to use AI could eventually be replaced.
Aditya Singh compared the issue to aviation and the development of autopilot. He cautioned that excessive automation could result in people losing the “muscle memory” needed when automation fails. He drew a parallel with legal education and warned that basic fundamentals of lawyering could be forgotten if young lawyers became excessively dependent on technology.
Nakul Dewan added that critical thinking and critical decision-making remained essential, comparing them to the pilot’s continued responsibility for the critical stages of take-off and landing.
Can arbitral institutions provide their own AI systems?
Another audience question considered whether Indian arbitral institutions or the judiciary could develop authenticated AI tools for use by counsel and arbitrators, potentially reducing reliance on open-ended public platforms.
Nakul Dewan said the question was directly connected to the issue of creating a level playing field. He referred to a recent discussion at a seminar in Singapore on whether arbitral institutions could themselves provide AI services.
He suggested that an institutional platform could, for example, give foreign counsel access to Indian legal authorities in an Indian-law arbitration while also providing a confidential environment in which pleadings could be processed. He said institutions should consider how much more technological infrastructure they could provide to make themselves more user-friendly.
Sameer Jain said developing institution-specific AI was technically feasible, although he emphasised that its usefulness would depend on the service the system was designed to provide.
Rishab Gupta said institutional arbitration in India could eventually incorporate clauses governing AI use and institution-provided technology capable of allowing confidential use by counsel and arbitrators. His larger concern, however, was maintaining India’s competitiveness when international arbitrations involved foreign counsel and large foreign law firms with substantially greater technological resources.
The possibility of AI tribunals and AI-assisted courts
The audience also raised the possibility of fully or predominantly AI-based adjudication.
Responding to a question about AI tribunals and courts, Aditya Singh pointed out that the AAA-ICDR’s AI Arbitrator was, at the time, the principal example discussed within arbitration. He reiterated that the pilot was restricted to a particular industry and document-only proceedings, with a human reviewer remaining “in the loop”. He said such a model could potentially become a way forward where parties agreed to it.
Sameer Jain noted that there was already debate about using AI to assist courts with large categories of relatively routine cases, including motor accident claims, traffic challenges and Negotiable Instruments Act matters. He also referred to the possibility of future developments in artificial general intelligence changing the capabilities of AI substantially.
Baiju Vasani, however, placed the issue in terms of human psychology. He said that, while younger generations might become increasingly comfortable with automated decision-making, people presently tended to prefer human decision-makers. He suggested that this preference would remain relevant even as AI capabilities advanced.
The emerging cost and access divide
The panel also considered whether advanced legal AI could widen the gap between large firms and smaller practitioners.
Rishab Gupta said one of the biggest concerns over the next five years would be the extent to which AI changed the competitive landscape. He observed that sophisticated systems such as Legora were expensive and generally built for large firms, with substantial investment in safeguards and infrastructure. His concern was that practitioners and smaller firms without comparable resources could be left behind.
The discussion nevertheless also considered the possibility that institution-provided technology could narrow some of these disparities by making advanced tools available to parties within an institutional framework.
AI’s environmental cost
An audience member raised another concern: the energy and water consumption associated with AI and data centres, and whether those costs would ultimately be borne by clients or addressed by law firms through corporate social responsibility initiatives.
Aditya Singh acknowledged that the energy and water demands of data centres had already become significant political issues. He said he expected pressure on AI companies to adopt carbon-offsetting measures or renewable energy solutions for data centres, although he did not suggest that a settled answer presently existed.
Nakul Dewan added that the environmental issue was not confined to the legal profession, noting that AI was also being widely used in medicine and other sectors and therefore required a broader societal response.
Closing reflections: preserving human intelligence
In the final exchanges, the speakers returned to the central tension between embracing AI and preserving professional judgment.
An audience member questioned whether lawyers who were not AI-savvy would increasingly be disadvantaged and whether younger lawyers might be pushed towards an expectation of near-infallibility. Sameer Jain responded that AI was unlikely to replace lawyers altogether, but lawyers who refused to use it could be replaced.
Aditya Singh returned to the aviation analogy, warning that sophisticated automation could create complacency and erode basic skills if professionals stopped practising them.
The discussion thus came back to a common theme across the panel: AI could substantially improve efficiency, assist with document-heavy work, identify patterns that human reviewers may overlook, and potentially transform aspects of arbitral administration and decision-making. But the speakers repeatedly distinguished using AI as an aid from delegating professional responsibility to it. For counsel, the emphasis remained on retaining command of the case and exercising independent judgment. For arbitrators, the discussion placed particular emphasis on personal decision-making, transparency where AI was used, and responsibility for the resulting award.
The session concluded with a broader reflection that the development of AI in arbitration would require not only technological adaptation but also attention to confidentiality, disclosure, equality of arms, professional training, institutional infrastructure and the preservation of the human judgment on which arbitral advocacy and adjudication continue to depend.
