If the first challenge of artificial intelligence in arbitration is understanding what the technology can do, the second is deciding who bears responsibility when it goes wrong. At LIDW 2026, the conversation moved beyond hallucinations and technical limitations to confront more difficult questions: how should confidential information be protected, what responsibilities do institutions and practitioners bear when using AI tools, and how might artificial intelligence reshape the economics and future structure of legal practice? It was these questions that dominated the second half of the session.
Confidentiality, Privilege and the Regulatory Lag
Few issues generated as much debate as the implications of AI for confidentiality and legal professional privilege.
The discussion was sparked by an audience member with a background in both chemical engineering and software programming, who questioned whether existing regulatory guidance adequately reflected the technical realities of modern AI systems.
In particular, he challenged recent guidance suggesting that lawyers risk breaching confidentiality simply by using AI tools with client documents. Such guidance, he argued, often assumes that documents are “uploaded” in a conventional sense when, in reality, information is processed through tokenised representations that differ significantly from traditional document sharing.
Mr. Charlie Morgan acknowledged the technical point but suggested that the more significant issue lies elsewhere.
Even if a document itself is not reproduced, interactions with AI systems generate new records in the form of prompts and outputs. Those records may themselves become relevant in future disputes. Lawyers, he observed, often interact with AI systems in an unusually candid manner. They speculate, brainstorm, test arguments and reveal assumptions. Many do so under the mistaken impression that they are communicating with an ephemeral digital assistant rather than creating potentially discoverable records.
“People are very candid with their AI bots. They think they are speaking to the ether when in fact they are speaking to a server that houses all their data.”
This reality may create entirely new categories of disclosure disputes.
Mr. Morgan noted that parties in litigation are already beginning to seek access to AI interaction logs, arguing that they may reveal admissions, strategic thinking or factual assumptions that never appeared in formal correspondence.
“We are going to have real challenges around this. There will be a lot of fights around discovery and production.”
Mr. Dmitri Evseev responded by drawing an important distinction between AI models and the applications that sit on top of them. One of the most common misconceptions about AI, he explained, is the belief that models learn continuously from every interaction.
He then clarified that in reality, most models are not retrained on user conversations in real time. Training occurs separately and periodically. The more significant issue therefore concerns the data retention policies of the software layer through which users access the model. For legal practitioners, understanding where information is stored, and for how long, may be more important than understanding the underlying model itself.
The discussion also turned to local models running entirely on a user’s own hardware. Here, Mr. Evseev adopted a straightforward position.
“If it is locally on your computer, it does not leave your computer. No third party ever sees it. I cannot see how it could possibly lead to a loss of privilege.”
The panel briefly discussed recent comments from the UK Immigration Tribunal suggesting that enterprise-grade systems such as Microsoft Copilot may present lower confidentiality risks than publicly available consumer tools. While the observation was acknowledged, there was general agreement that the jurisprudence remains in its infancy and that definitive answers have yet to emerge.
Institutional Responses: Rules in a Shifting Landscape
The discussion then moved from individual practitioners to the institutions that shape international arbitration. How, Ms. Limond asked, are arbitral institutions responding to AI? And perhaps more importantly, should they be responding at all?
Mr. Dmitri Evseev was uniquely positioned to address the question. As one of the architects of the Silicon Valley Arbitration and Mediation Centre AI Guidelines1, he had firsthand experience of attempting to formulate institutional guidance in an area evolving at extraordinary speed.
What surprised him most, as he stated, was not the complexity of the task but the absence of engagement from arbitral institutions. After a nine-month public consultation process, he reported, not a single official or unofficial comment was received from any major arbitral institution. He suggested that the explanation may be partly strategic as institutions are understandably reluctant to bind themselves to detailed rules that risk becoming obsolete almost immediately. Technological developments are occurring so rapidly that any rigid framework could be outdated before implementation and the result is a degree of institutional caution.
“Arbitral institutions don’t want to tie themselves down to somebody else’s rules. And ideally, perhaps, they don’t want to get too involved in this because it is very murky waters.”
Mr. Charlie Morgan offered insight from the perspective of the ICC AI Task Force, which is expected to publish recommendations later in 2026. Even within a relatively compressed timeframe by institutional standards, he noted, the technology will have evolved significantly before the work is completed. Consequently, many institutions appear to be gravitating toward principle-based approaches rather than technology-specific regulation.
The emerging consensus focuses on familiar obligations:
-
lawyers must stand behind their submissions,
-
fabricated authorities remain unacceptable,
-
and arbitrators must exercise independent judgment.
The discussion eventually turned to one of the more controversial developments in the field: AI-assisted arbitrators.
Certain initiatives have experimented with models in which AI systems produce draft decisions for subsequent human review. While the concept attracted interest, the panel approached it with caution. Mr. Morgan’s assessment was measured but sceptical.
“The market is probably not quite ready for AI arbitrators. And I think the technology is probably not quite ready for it either.”
He expressed that the idea may one day become commonplace, for now, however, the profession appears more comfortable using AI to assist human decision-makers than to replace them.
What Clients Actually Want, And the Billing Paradox
As the discussion moved from technology and regulation to commercial realities, the focus shifted to the stakeholders ultimately driving change in legal services: clients.
Mr. Stéphane Altounian highlighted what he described as a fundamental tension within the market. On one hand, clients expect legal work to become cheaper as AI adoption increases. On the other, firms are making substantial investments in technology licences, infrastructure, security, training and implementation.
“These are completely conflictive incentives. It is hard to know where things are going to go.”
Mr. Altounian expressed that traditional assumption is that technology reduces work, yet several panellists suggested that the opposite may often be true.
Mr. Morgan observed that AI is frequently generating more activity rather than less. He stated that clients now arrive with more information, more questions and more detailed instructions than ever before. They are better informed, more inquisitive and increasingly willing to challenge advice. Lawyers, in turn, must devote additional time to analysing, contextualising and responding to those questions. He added that the proliferation of AI-assisted self-represented parties has added another layer of complexity.
He pointed out that individuals who previously lacked the resources to generate lengthy submissions can now produce extensive arguments with the assistance of AI tools. While many of those arguments may ultimately lack merit, they nevertheless require careful review and response.
Mr. Morgan noted that this phenomenon can be particularly challenging in arbitration. Unlike courts, arbitral tribunals are sometimes less inclined to summarily dismiss weak arguments, creating situations in which parties and counsel must devote significant resources to addressing machine-generated submissions that would previously never have existed.
The discussion revealed a broader truth about technological progress in dispute resolution, that the volume of legal work is not shrinking, it is expanding. The amount of information that lawyers must process continues to increase exponentially, and AI is enabling parties to generate and analyse material on an unprecedented scale.
Mr. Altounian identified what may prove to be the most significant long-term challenge to the conventional law firm model. A growing number of sophisticated corporate clients are openly discussing the possibility of bringing large portions of legal work in-house through AI-assisted teams.
“Some corporate clients are waiting for just the next step of AI where they will be able to do 80% of the work internally with AI tools and then only consult a high-profile partner for orchestrating the strategy of the case.”
This is not merely a theoretical future.
He expressed that several large organisations are already exploring how AI might allow them to reduce reliance on external counsel for routine research, document review, drafting and analysis while retaining specialist lawyers for strategic decision-making, advocacy and risk management. The implications for legal services could be profound. Yet even as technology reshapes workflows, the discussion repeatedly returned to a central theme that complexity is not disappearing.
One audience member described a recent e-discovery exercise in which generative AI reportedly saved approximately 300 days of lawyer time on a single matter. The figure was striking but the broader lesson was arguably even more important.
Technology is not reducing the amount of information available to lawyers, it is simply making it possible to process vastly larger quantities of material more effectively. That observation led to one of the most memorable exchanges of the session.
As one participant observed, there is little value in parties submitting hundreds of pages of machine-generated argument if decision-makers ultimately rely on technology to reduce those submissions to a few digestible pages. The resulting asymmetry is unlikely to remain sustainable.
Whether through page limits, procedural reforms or technological adaptation, arbitration will eventually need to find a balance. As the moderator remarked, the existing inequality of arms created by machine-generated volume “is not going to continue.”
The Right Way to Use AI: Grounding, Verification and Context
As the session drew toward its conclusion, a notable consensus emerged among the panellists. Despite differences in perspective and professional background, all agreed on one fundamental principle: the value of AI lies not in replacing legal judgment but in enhancing access to information.
Throughout the discussion, the most persuasive arguments in favour of AI centred on retrieval rather than generation. Mr. Mark Feldner repeatedly returned to what might be described as a data-first philosophy.
For him, the most valuable use of advanced AI systems is not their ability to produce answers but their ability to locate relevant information quickly within trusted datasets.
“What you ultimately want to rely on is the document the model has found, and the source information.”
His emphasis was on verification, that lawyers should trust source materials, not machine-generated prose. He further stated that the AI’s role is to identify relevant information, organise it effectively and present it in a manner that facilitates human analysis.
Mr. Dmitri Evseev framed the issue in slightly different terms. He suggested that the legal profession is gradually moving beyond the era of “prompt engineering” and into what he described as “context engineering.” The distinction is significant.
He reasoned that for much of the AI revolution, attention has focused on crafting increasingly sophisticated prompts. Yet, he argued that the more important challenge lies elsewhere. He emphasised that the real question is not how to ask better questions but how to ensure that AI systems have access to precisely the right information. And only that information.
“If you are asking a question about a specific matter, it should have no information about the next matter that you are working on, the other matter for another client.”
Managing context effectively, he suggested, may ultimately prove more important than mastering prompts.
For arbitration practitioners dealing with confidential information across multiple matters, he emphasised that context management is not merely a technical issue, rather it is a governance issue.
Mr. Evseev, offering a succinct articulation of AI’s proper role in dispute resolution said,
“The fundamental problem with the use of AI in arbitration is that you are essentially leveraging something that is outside the record, outside the pleadings, outside the evidence. You can treat the AI answer like going down the hall and asking a smart colleague, but you would not cite that in a brief.”
The analogy resonated with many in the room. The moment an AI system helps a lawyer identify a relevant authority, argument or piece of evidence that would otherwise have remained hidden, it delivers value that transcends simple time-saving.
Conclusion: A Defining Moment for Arbitration
The panel did not attempt to predict the future of artificial intelligence in arbitration with certainty. Indeed, one of the discussion’s defining characteristics was its refusal to indulge in either utopian optimism or technological pessimism. No panellist claimed to know exactly where AI would take the profession. What emerged instead was something arguably more useful, an honest and experience-driven assessment of the choices confronting arbitration practitioners today.
The session’s title, The Human Fight Back in Arbitration, ultimately proved somewhat misleading. The discussion was not about resisting artificial intelligence but about resisting complacency. The real challenge facing the profession is not technological displacement but technological competence.
Lawyers, arbitrators, institutions and clients must develop the skills necessary to understand these systems, deploy them responsibly, verify their outputs and govern their use effectively.
Throughout the discussion, the panellists repeatedly returned to the qualities that remain distinctly human: judgment, credibility, nuance, accountability and strategic thinking.
As Mr. Charlie Morgan observed toward the close of the session, successful use of AI requires active engagement rather than passive reliance.
“You do the hard thinking. You think, what should this output look like? Then you guide the tool as to what it should be and then you check the output. In that way you can save 10, 20, 30, 40, 50% of time on a task without compromising the output.”
This observation captures the central lesson of the discussion that artificial intelligence is neither a replacement for legal judgment nor a threat to it, it is a tool.
The question confronting international arbitration is not whether that evolution will occur, it is whether the profession will evolve alongside it, or find itself overtaken by the very technology it seeks to harness.
1. SVAMC Guidelines on the Use of Artificial Intelligence in Arbitration, published April 30th, 2024
