THE AUTHOR:
Santiago Rodríguez, Senior Associate at Arias SLP & Adjunct Professor at IE Law School
While established doctrine exists regarding the applicable boundaries to tribunal secretaries and the assistance they may provide arbitrators, no equivalent framework exists for arbitrators’ use of artificial intelligence (“AI”). Through practical examples, this article seeks to draw the line between permissible and impermissible AI use by arbitrators.
Conflicts of Interest, Disqualifying Facts and Non-Disqualifying Facts
Before examining specific examples of permissible and impermissible uses of AI by arbitrators, we must properly label and define the potential issues that may arise in this context.
Conflicts of interest are defined by the IBA Guidelines on Conflicts of Interest in International Arbitration 2024 (“IBA Guidelines”) as:
- any situation that may cause the arbitrator to doubt their ability to be impartial or independent; or
- facts or circumstances that, from the point of view of a reasonable third person, would give rise to justifiable doubts as to the arbitrator’s impartiality or independence.
Conflicts of interest must therefore necessarily affect arbitrators’ impartiality or independence. It is not apparent how the use of AI, by itself and without additional context, may create a conflict of interest. Therefore, we shall avoid using this term when addressing potential issues that may arise regarding the use of AI by arbitrators.
It seems more appropriate to label AI use by arbitrators as either Disqualifying Facts or Non-Disqualifying Facts.Disqualifying Facts are those that may lead to the disqualification of an arbitrator or annulment of an award for reasons unrelated to conflicts of interest, such as incapacity, failure to act without delay, or conducting the procedure in breach of the parties’ agreement. Non-Disqualifying Facts are those that, although they may relate to Disqualifying Facts, do not cross the threshold necessary to disqualify an arbitrator or annul an award.
Disqualifying Facts are addressed in different rules and laws:
- Article 14.1 of the ICC Arbitration Rules 2021 allows challenges to arbitrators for “an alleged lack of impartiality or independence, or otherwise.” This “otherwise” encompasses Disqualifying Facts.
- The UNCITRAL Model Law on International Commercial Arbitration 2006 also separates the grounds for challengingarbitrators based on reasons relating to their independence and impartiality (Article 12.2) from other reasons, such as a failure or impossibility to act without delay, if he withdraws from his office or if the parties agree on the termination (Article 14.1).
- Article 34.2.a.iv of the UNCITRAL Model Law provides that awards may be set aside if the arbitral procedurewas not in accordance with the parties’ agreement or with the law.
- Article 34.2.b.ii of the UNCITRAL Model Law provides that an award may be set aside if it is contrary to the public policy of a State.
Examples :
Once we agree on the applicable vocabulary to address the potential issues that may arise regarding AI use by arbitrators, we can now focus on concrete examples to draw the line between appropriate and inappropriate AI use by arbitrators.
Example 1
Does the use of AI by arbitrators, by itself and without additional context, (i) create a Disqualifying Fact or a Non-Disqualifying Fact, and (ii) does it require disclosure? (Although disclosure usually refers to the arbitrator’s duty to reveal any fact or circumstance which may, in the eyes of the parties, give rise to doubts as to the arbitrator’s impartiality or independence, for the purposes of this article we will refer to disclosure in broader terms to include the duty to reveal Disqualifying Facts as well.)
AI encompasses a broad spectrum of tools, ranging from basic spellcheck and translation software to sophisticated large language models. It would be difficult to assert that the mere use of AI, without additional context, is sufficient to disqualify an arbitrator or challenge the enforcement of an award. Moreover, it would be impractical and overlyburdensome for arbitrators to disclose the use of any AI tool for any routine task.
Therefore, the use of AI by arbitrators, by itself and without additional context, constitutes a Non- Disqualifying Fact.However, with additional context, it may evolve into a Disqualifying Fact or may require disclosure, as will be explained in the following examples.
Example 2
Does the use of AI by arbitrators for drafting (i) create a Disqualifying Fact or a Non- Disqualifying Fact, and (ii) does it require disclosure?
It is commonly accepted – and reflected in both the Terms of Reference and procedural orders – that tribunal secretaries may assist arbitrators in performing mechanical tasks when drafting procedural orders or awards. Such tasks include, for example, drafting the parties’ background, the procedural history of the arbitration, or performing translations.
AI can perform some of these mechanical tasks. It would be difficult to sustain that using AI to execute such mechanical tasks constitutes, by itself and without additional context, a Disqualifying Fact. This is particularly so where such use is specifically permitted under the Terms of Reference or a procedural order. For this reason, it appears unnecessary for an arbitrator to disclose the use of AI for drafting, provided it involves only mechanical tasks, and this has been accepted by the parties.
Without additional context, this example illustrates a Non-Disqualifying Fact. However, depending on the specificcircumstances, a similar scenario may evolve into a Disqualifying Fact or may require disclosure, as will be explained in the following example.
Example 3
Does an arbitrator’s reliance on AI for deciding and drafting a decision (i) create a Disqualifying Fact or a Non-Disqualifying Fact, and (ii) does it require disclosure?
Consider an arbitrator who relies on AI to decide a case. Even if the decision is error-free and unanimously signed bythe tribunal, could the losing party successfully challenge the award upon discovering the AI’s role?
While it is possible that the arbitrator would have reached the same result as the AI and that such assistance did not alter the outcome, the fundamental problem is that the arbitrator delegated the decision-making process, thereby violating the mandate provided by the parties. The arbitration agreement entrusts the responsibility of deciding the case to thearbitrator, not to AI. A breach of this mandate could constitute grounds for annulling an award, either because the procedure was not conducted pursuant to the parties’ agreement or because it contravenes the public policy of the State – depending on the applicable law in each case. Even if such delegation were disclosed by the arbitrator, it would still constitute a clear breach of mandate.
Therefore, this scenario constitutes a Disqualifying Fact. However, by tweaking this same example, the answer may change.
If the arbitrator decides independently, outlines the decision, and delegates only the drafting of a first draft to AI,following the arbitrator’s instructions and under the arbitrator’s supervision, does this (i) constitute a Disqualifying Fact, and (ii) require disclosure?
Since decision-making remains with the arbitrator and only the drafting is delegated, no apparent Disqualifying Fact arises. It would need to be demonstrated that delegating the drafting of the decision violates the arbitrator’s mandate. This could occur where the parties have explicitly required arbitrators not to delegate this task or to disclose the use ofAI for any task.
This issue is controversial. It is common practice to state in the Terms of Reference or procedural orders that tribunal secretaries may assist arbitrators in drafting the award under the tribunal’s supervision and in accordance with its instructions. The same boundaries could apply to AI.
Moreover, as explained above, AI is a broad term encompassing a wide range of tools. Requiring arbitrators to disclose the use of AI for any task would be unduly burdensome and potentially unenforceable. This scenario raises a distinct andmore complex question: whether an arbitrator’s failure to disclose the use of AI, if required by the parties’ agreement,constitutes a Disqualifying Fact, even when the AI is used for mechanical tasks.
Absent an agreement requiring arbitrators not to delegate the drafting of decisions or to disclose the use of AI generally,it would be difficult to argue that the preparation of a first draft, following an outline provided by the arbitrator andconducted under the arbitrator’s supervision, constitutes a Disqualifying Fact. In such circumstances, there is no apparent breach of the arbitrator’s mandate, as decision-making has not been delegated and no party agreement prohibits this practice. By itself, this scenario appears to constitute a Non-Disqualifying Fact that does not require disclosure.
Conclusion
The practical reality is that expecting arbitrators to personally review every document and draft every word in every procedural order or decision is unrealistic. Arbitrations are becoming increasingly complex, involving longersubmissions, more documents, and additional procedural hurdles. Assistance from AI can actually enhance arbitrators’performance by allowing them to delegate mechanical tasks, thereby providing more time to focus on the key issues of the case.
The arbitrator’s mandate – grounded in the arbitration agreement or in law – requires arbitrators to act diligently andresolve the dispute for which the parties appointed them. Delegating decision-making to AI would in principle constitute a breach of this mandate and constitute a Disqualifying Fact. However, when it comes to non-decisional and mechanicaltasks, the line becomes less clear and must be analysed on a case-by-case basis.
As explained above, there are instances where the mere use of AI for mechanical tasks, without additional context, is permissible and should not, by itself, constitute a Disqualifying Fact. However, if the parties’ agreement explicitly forbids arbitrators from delegating the drafting of any decision or requires them to disclose the use of AI broadly, then this same fact may evolve into a Disqualifying Fact because it could be considered a breach of their mandate.
For the above reasons, we propose that arbitration institutions and organizations provide clear guidelines to assist arbitrators in drawing the lines between permissible and impressible AI use by arbitrators. This way, the arbitration community could advance more smoothly towards the proper adoption of AI tools.
ABOUT THE AUTHOR
Santiago Rodríguez is a Senior Associate at Arias SLP in Madrid, with nearly a decade of experience in commercial and investment arbitration. His practice focuses on complex disputes, particularly in the energy, oil and gas, and construction sectors. In addition to his practice, Santiago serves as an Adjunct Professor at IE Law School, where he teaches courses on the philosophy of law and international dispute resolution. He is admitted to practice in Venezuela and Spain, and is currently pending admission to the New York Bar.
*The views and opinions expressed by authors are theirs and do not necessarily reflect those of Arias SLP, its employees, or its clients, or Daily Jus, Jus Mundi, or Jus Connect.
