by Andrew Flake
In Kashuo Ishiguru’s novel “Klara and the Sun,” the eponymous character is an AF, or artificial friend, a sophisticated, artificially-intelligent robot designed to be a companion for human children. Klara is kind, surprisingly observant and intuitive, and even heroic in the ways she communicates and supports Josie, the child whose family purchases Klara. But what also struck me as I read, and what the book develops so well, is the ways that despite her kindness, Klara sees and understands the world in some fundamentally different ways — notably, because she is solar-powered, in viewing the Sun as a kind of deity, one with special powers of healing. I won’t give away anything, and will say instead that part of the wonder of the book is how Ishiguro develops this picture of Klara through all of the individual observations she makes as she leaves a storefront window and begins to navigate the world. She draws conclusions and reasons in ways that a human never would.
In much the same way, artificial intelligence in today’s frontier models — Claude Fable, Gemini, GPT 6 — despite our tendency to ascribe humanlike qualities to them, and their ability to exhibit many humanlike qualities, remain a mystery in many ways to even their designers in the AI labs. How they reason, how they arrive at decisions, and how they interact as agents in groups, are not entirely predictable. This phenomenon, by itself, is one prudential reason for legal professionals relying on AI, to be cautious in the use cases we choose.
I think that is especially so when we consider the ways in which neutrals, mediators and especially arbitrators, might employ AI. During a recent discussion of these issues with other arbitration practitioners at this year’s excellent AtlAS Conference in Atlanta, I made the comment to a friend that the question with AI is not just “How might I use it,” because the range of capability continues to expand, but also, from the standpoint of our ethical rules and norms, “May I use it,” and lastly, the personal and often overlooked question, “Do I want to use it?” This last category involves areas that we enjoy, for example, writing or perhaps doing our own travel research for a vacation, that we derive pleasure and benefit from and may not want to supplant with AI.
On the first topic, how arbitrators use AI, and the second, whether we may permissibly do so, it strikes me that the role of a tribunal secretary provides some useful analogues. Most familiar to international arbitration practitioners, the tribunal secretary is an assistant to the arbitrators, selected in consultation with the parties, who supports the tribunal in defined and non-deliberative ways. The secretary is bound by the same ethical rules, makes disclosures to the parties, and works under the tribunal’s supervision through the case on tasks that, to quote ICDR’s guidelines for tribunal secretaries, “increase the efficiency and [] reduce the aggregate cost” of the arbitration. Critically, the tribunal secretary lacks any decision-making authority.
The 2026 update to the ICC’s Rules, responding to cases in which the tribunal secretary was viewed to have overstepped these limits, provide specific constrains along the same lines: Article 44 provides that “[a]fter consulting with the parties, the arbitral tribunal may appoint a tribunal secretary to work under the arbitral tribunal’s direction and control, without delegating its decision-making authority” and that tribunal secretaries must “satisfy the same independence, impartiality and confidentiality requirements as arbitrators.”
Our current ethical framework, including the duties to exercise independent judgment and of non-delegation in Canon V of the Code of Ethics for Arbitrators in Commercial Disputes, and institutional guidelines on AI use, including from AAA-IDCR, SVAMC, and the Chartered Institute of Arbitrators, gives us solid scaffolding for using AI responsibly to capture benefits, notably cost and time savings, while protecting the essential attributes of arbitration. But we are going to need to stay vigilant, and to continue to adapt those guidelines, as AI capabilities continue to accelerate and — in the case of agentic AI — move from mere tools to technology with autonomous functions. As we think about how to do so, our historical use of tribunal secretaries and the rules we’ve developed around their role should be helpful. ABF