TOOLS OF THE TRADE – NOVEMBER 2026

THE ARBITRATOR, THE ALGORITHM AND THE CONFIDENTIAL QUESTION[1]

INTRODUCTION

Chris Binnington explores the tension which arises as a result of the use of Artificial Intelligence (AI) techniques in construction dispute resolution by third-party neutral.  In particular, the problems of confidentiality when dispute submissions or aspects of the dispute process are uploaded into a Large Language Model (LLM).

THE PROBLEM DEFINED

The rapid emergence of AI tools has prompted understandable questions within the dispute resolution community.  Arbitrators, adjudicators and dispute board members increasingly encounter AI-assisted technologies capable of summarising documents, analysing submissions, generating drafts and organising complex information.  As adoption grows, attention is shifting from what AI can do to how it should be used by those entrusted with making independent decisions.

The starting point is that the legitimacy of any adjudicative process depends upon the neutrality, independence and personal judgment of the decision-maker.  Parties appoint a particular individual because of that person’s expertise, experience and capacity for independent reasoning.  These responsibilities cannot be delegated to a machine.

At the same time, it would be unrealistic to ignore the practical benefits that AI can offer.  Modern disputes frequently involve thousands of pages of pleadings, witness statements, expert reports, correspondence and contractual documentation.  AI tools may assist by organising information, identifying inconsistencies, preparing chronologies, summarising arguments, improving drafting quality and checking cross-references.  Used appropriately, such tools can enhance efficiency without compromising the integrity of the decision-making process.

The distinction lies in the nature of the assistance being provided.  There is a significant difference between using AI as an administrative or analytical aid and allowing it to influence substantive findings.  Assistance that resembles the work of a research assistant, law clerk or editorial reviewer may be entirely appropriate.  By contrast, reliance upon AI to determine disputed facts, assess witness credibility, formulate legal conclusions or introduce arguments not advanced by the parties presents much greater concerns.

The key principle is that the human decision-maker must remain the author of the decision in every meaningful sense.  All findings of fact, legal conclusions and exercises of discretion must be independently evaluated and adopted by the appointed neutral.  AI-generated content should be treated as a draft for review, not as an authoritative source.

Questions also arise concerning confidentiality.  Many practitioners are understandably cautious about entrusting confidential submissions, witness statements, expert reports and draft decisions to third-party AI systems.  A common misconception is that information uploaded to an AI platform automatically becomes publicly available.  In general, information submitted to such systems is not accessible to other users of the platform or searchable by members of the public.  However, this does not end the inquiry.  Information may still be stored, processed and retained within the provider’s systems in accordance with applicable contractual terms, privacy policies and legal obligations.  The precise treatment of data may differ materially between consumer, professional and enterprise offerings.

Accordingly, dispute resolution professionals should not assume that confidentiality obligations are satisfied merely because information is inaccessible to other users.  Consideration should also be given to where data is stored, the provider’s retention policies, whether uploaded material may be used to improve services, the availability of opt-out mechanisms and the legal framework governing disclosure obligations in the relevant jurisdiction.

THE POTENTIAL SOLUTION

The prudent course is to approach AI-assisted work with the same degree of care that would be applied when engaging external consultants, cloud-based document management systems or third-party transcription services.  Before uploading confidential materials, decision-makers should satisfy themselves that the applicable terms of use, security protections and data-handling arrangements are compatible with their professional and contractual obligations.

THE CONFIDENTIALITY PROBLEM

A further issue concerns the potential discoverability of AI-assisted work product.  While communications with AI systems are generally private and not accessible to opposing parties merely because they exist, users should avoid assuming that such records can never become relevant in subsequent proceedings.  Depending upon the applicable legal framework, court order, arbitral ruling, regulatory investigation or professional disciplinary process, questions may arise concerning the existence and relevance of AI-generated drafts, prompts, analyses or research undertaken during the decision-making process.

The position may, however, be different in relation to the private working papers, notes, drafts and deliberative materials of the decision-maker.  South African authority has recognised an important distinction between the formal record of proceedings and the internal deliberative process of the decision-maker.  Courts have acknowledged that an arbitrator’s notes do not ordinarily form part of the formal record and that the integrity of independent decision-making may require protection of the deliberative process itself.

This distinction raises an interesting question in the context of AI.  It may ultimately be necessary to distinguish between AI-generated material forming part of the evidential or procedural record, on the one hand, and AI-assisted drafts, research notes, working memoranda or deliberative analyses generated solely for the decision-maker’s internal use, on the other.  Whether existing principles concerning arbitral notes and draft awards will extend to AI-assisted working materials remains an issue that has yet to be fully explored by the courts.

The issue is not merely one of discoverability.  A prior question may be whether the arbitrator, adjudicator or dispute board member was entitled, under the applicable arbitration agreement, adjudication rules, institutional rules, confidentiality undertakings or professional obligations, to upload the material to a third-party AI service in the first place.

Prudent practitioners may therefore wish to adopt clear safeguards.  These might include independently verifying all factual and legal conclusions, maintaining personal control over the reasoning process, avoiding reliance on AI-generated authorities without verification, ensuring that confidential information is handled appropriately and exercising caution before uploading sensitive material.

THE CONCLUSION

The debate surrounding AI in dispute resolution is still developing.  What is already clear, however, is that the critical question is not whether AI is used but rather whether its use preserves the independence, impartiality and accountability of the decision-maker.  So long as those principles remain paramount, AI can serve as a valuable tool without undermining confidence in the dispute resolution process.

Chris Binnington
Director, Rules Committee Chair
ASSOCIATION OF ARBITRATORS (SOUTHERN AFRICA) NPC


Missed a Previous Arbitrarily Speaking Issue?

Catch up here.


[1]

1 September 2026.