Emerging guidelines allow the use of artificial intelligence as an assistive tool in researching, organizing information, and testing arguments, but they emphasize that the arbitrator must retain full responsibility for analysis and the final decision.
For instance, an arbitrator may use AI to identify legal references or review counterarguments to an initial interpretation, provided that the results are independently reviewed and verified.
However, requesting to determine which party has the stronger argument, or to report on the entitlement to compensation, or to suggest how to resolve an objection regarding jurisdiction, approaches delegating part of the dispute resolution task.
The guidelines issued in 2025 by the Chartered Institute of Arbitrators ("Ciarb") assert that arbitrators may not relinquish their decision-making authority in favor of AI.
They also call on arbitrators to verify the information provided by these systems and to take responsibility for all aspects of the decision.
The Silicon Valley Arbitration and Mediation Center ("SVAMC") adopts a similar stance, prohibiting the delegation of personal powers of the arbitrator to an AI tool, emphasizing that these tools cannot replace their independent analysis of facts, law, and evidence.
The use of AI raises additional risks related to the confidentiality of arbitration proceedings, which often involve sensitive contracts, pleadings, witness statements, expert reports, and business information.
Uploading these materials to unsecured tools may lead to retaining or using them in training or accessing by third parties.
Therefore, the guidelines recommend reviewing data storage and protection policies before inputting any confidential information.
Generative AI can also produce non-existent judicial references or incorrect legal conclusions framed in a plausible manner.
Thus, the arbitrator remains responsible for returning to original sources and verifying the validity of the rules and principles relied upon.
Another issue arises when AI suggests an argument or reference not presented by either party to the dispute. If this element becomes influential in the outcome, it should be presented to the parties and they should be given an opportunity to comment, rather than being included in the decision without their knowledge.
Current guidelines do not adopt a uniform position on disclosing the use of AI.
The "Ciarb" guidelines encourage arbitrators to consult with the parties before using these tools, while "SVAMC" guidelines do not impose a general disclosure but require it when the arbitrator intends to rely on information generated by AI outside the case file.
The American Arbitration Association ("AAA") has launched an optional procedure for eligible disputes, allowing AI to analyze party briefs and prepare a draft decision, which is then reviewed by a human arbitrator who may modify or rephrase and finalize the decision.
However, this procedure is based on explicit consent from both parties to the dispute.
The repercussions of uncontrolled AI use are beginning to appear before the courts. In the case of "LaPaglia v. Valve Corporation," one party claimed that the arbitrator delegated their judicial role to AI, but a U.S. court dismissed the amended petition in December 2025 on jurisdictional grounds, without addressing the substance of the claim.
In April 2026, the Quebec Supreme Court annulled a decision after it was found that non-existent references and principles were central to the arbitrator's reasoning, and that AI use occurred without adequate verification and included an improper delegation of part of the dispute resolution task.
The court did not consider mere use of AI as a reason for annulment; rather, the issue lay in reliance on fabricated information and failure to verify it, breaching agreed-upon procedures.
Thus, the dividing line does not concern the arbitrator asking AI for its opinion but rather the nature of the task performed and how to handle its outputs.
AI can assist in research and testing analysis, but it should not replace the independent judgment of the arbitrator or affect the integrity of arbitration procedures.