CPR's AI in Arbitration Conference Explores Governance, Ethics, and the Tools Themselves

The International Institute for Conflict Prevention and Resolution—CPR held a daylong hybrid conference on artificial intelligence in arbitration, on Sept. 15 at White & Case's New York offices, along with an online audience, brought together arbitrators, advocates, in-house counsel, and a Microsoft product team.

The event consisted of four sessions on how AI is reshaping the practice, and how far the rules governing that practice have (and haven't) caught up.

The program, presented by CPR Dispute Resolution Services, titled "The AI-Ready Neutral: Tools for Application, Assessing Risks, and the Future of AI," was designed to equip neutrals and advocates to engage with artificial intelligence thoughtfully, practically, and responsibly.

Participants began with an overview of the evolving AI landscape, including key principles, legal developments, current applications, and emerging trends in arbitration; moved into a hands-on demonstration of Microsoft’s AI client, Copilot, and effective prompting techniques led by presenters from Microsoft; examined governance, ethics, risk assessment, and common pitfalls specific to neutral practice.  The conference closed with a keynote on where AI is headed and what its continued development may mean for dispute resolution.

Panel 1: Microsoft Walks Through 365 Copilot

The day opened with a live demonstration from Jake Thornburgh, a senior paralegal on Microsoft's litigation team, and Nathan Haley, Microsoft's principal financial discovery manager—both based at Microsoft’s Washington state headquarters--who introduced three tiers of Microsoft 365 Copilot that they called, jokingly, "mild, medium, and spicy."

Much of the audience's attention went to two concerns, cost and confidentiality.

Haley explained that Copilot Cowork, is billed on consumption, prompting extended questioning from arbitrators in the room about how to anticipate a monthly bill.

On confidentiality, Thornburgh pointed to a feature called Work IQ, which controls whether Microsoft 365 Copilot searches a user's own emails, Teams messages, and files when generating a response.

One attendee, an arbitrator managing multiple confidential matters, asked directly whether she could restrict the tool to a single case's folder. "Yes," Haley said. "You control the information that it has access to and what you would like it to consider in its responses."

* * *

Panel 2: Counsel and Arbitrator Perspectives on AI in International Practice

The second panel paired Myrna Barakat Friedman, an independent arbitrator who heads her own New York firm, BarakatADR, and a member of CPR's Panel of Distinguished Neutrals who chairs a New York City Bar Association AI and Arbitration Subcommittee, with Sven Volkmer, a New York partner in White & Case's International Arbitration practice.

The two speakers described divergent approaches to disclosure. Some arbitrators, Friedman said, now include a standing AI-use disclosure in their arbitration profile, giving parties a window to object before a matter proceeds; if even one party objects, the practice is generally not to use it in that case.

She was candid about the discomfort even that process can create. Describing one matter where she raised AI use with her co-arbitrators before even approaching the parties, Friedman recalled, "I got so many questions from the co-arbitrators, I'm like, I give up. I'm not even going to try with the parties."

Both speakers said they are far more comfortable using AI in domestic U.S. matters than in international ones, citing uncertainty over how courts in the European Union or China might treat an award informed by AI-assisted research.

* * *

Panel 3: Ethics and Governance

The afternoon panel split between San Diego-based Cynthia Cwik of JAMS, who addressed the broader governance landscape, and Thomas White a litigation partner in the New York and Washington, D.C., offices of Sullivan & Cromwell, who worked through the professional conduct rules he sees as directly implicated by AI use in practice.

Cwik opened with public statements from AI industry leaders, including Anthropic CEO Dario Amodei on how little even AI's own creators understand about how their models work, and OpenAI's chief scientist on the approach of "recursive self-improvement," in which AI systems begin materially improving their own code, before turning to concrete governance developments.

She reviewed the EU AI Act's tiered, risk-based approach, which restricts but does not ban AI use in judicial and quasi-judicial settings; the absence of comprehensive federal AI regulation in the U.S.; and California's SB 574, which would bar arbitrators from delegating decision-making to AI and attorneys from delegating the practice of law to it (California Gov. Gavin Newsom has until Sept. 30 to sign or veto the bill; if there's no action, it automatically becomes law). 

White, who is co-head of Sullivan & Cromwell's International Arbitration and Global Dispute Resolution Practice, then walked through several New York Rules of Professional Conduct, opening with a deliberately counter-intuitive point. The duty of competence, he suggested, may create an obligation to use AI, not merely a duty to guard against its misuse.

"If someone is stubbornly refusing to use the technology without a good reason," he said, "that could be an ethical issue, I would suggest."

He traced how the duty to communicate with clients extends to explaining a firm's use of AI, how the rule against unreasonable fees raises open questions about passing AI costs on to clients, and how the duty to supervise nonlawyers extends naturally to supervising AI output.

* * *

Featured Talk: Simulation, Personas, and the Future of Legal Practice

The day closed with a keynote from Megan Ma, Executive Director of Stanford Law School's Legal Innovation through Frontier Technology Lab (LIFT Lab) and a member of CPR's Board of Directors.

She opened with the story of a viral proof of the Jacobian conjecture-- a famous problem in algebraic geometry--produced with the help of AI model Claude by a young mathematician. The mathematician, it turned out, had studied under two Fields Medalists—scholars who have earned the International Mathematical Union’s highest honor. She used the story of the viral proof to argue that AI's recent breakthroughs in mathematics are sharpening specialists rather than replacing them.

She balanced that optimism against a petition from mathematician Terence Tao warning that frontier AI labs often train models to optimize for correct final answers, while missing what he called the "art and magic" of the discovery process itself, a caution she said applies directly to legal work.

Ma traced five inflection points she believes moved AI models from novelty to genuine collaborator. These included the leap in common-sense reasoning with OpenAI's GPT-4, the arrival of "reasoning" models that deliberate before answering, coding agents maturing into agentic collaborators, models learning to ask clarifying questions rather than guess at intent, and, most recently, tools that actively probe a user's preferences rather than simply executing instructions.

* * *

Across all four sessions, a few themes recurred. Competent, ethical practice may now require engaging with AI rather than merely guarding against it. Responsibility for verifying AI output remains squarely with the human professional. And no one in the room, not the arbitrators, the advocates, or the ethics specialists, has yet settled on where "using a tool" ends and "delegating judgment" begins.

* * *

A deeper diver into the event,  "The AI-Ready Neutral: Tools for Application, Assessing Risks, and the Future of AI," will appear in the November issue of Alternatives to the High Cost of Litigation.

* * *

The author, a third-year law student at the Northeastern University School of Law in Boston, is a CPR Institute fall intern.

[END]