AI in Arbitration: Efficiency Tool or Grounds to Challenge an Award?

Artificial intelligence (AI) is quickly becoming part of the legal process, including in arbitration. Used carefully, AI can help lawyers, parties, and neutrals work more efficiently, organize information, and improve the presentation of a case. Used carelessly, however, AI can raise serious questions about reliability, transparency, and whether a decision-maker has improperly delegated judgment that must remain human.

The American Arbitration Association International Centre for Dispute Resolution has responded to these concerns by publishing the AAAi Standards for AI in Alternative Dispute Resolution (ADR). Those standards identify six core principles for AI use in ADR: ethical and human-centered values; privacy and security; accuracy and reliability; explainability and transparency; accountability; and adaptability. The standards apply across the ADR process, including to administrators, arbitrators, mediators, and the parties who appear before them.

The standards are an important step, but they do not answer every practical question. In particular, they do not clearly explain what happens when an arbitrator allegedly misuses AI, how a party would prove that misuse, or whether the remedy would be disqualification, correction, or vacatur of an award. That gap matters because arbitration rules generally provide only limited mechanisms to revisit an award once it has been issued.

Under several AAA rules, an arbitrator may be disqualified before an award is entered for issues such as lack of independence, inability or refusal to perform duties diligently and in good faith, or other grounds supplied by applicable law. After it is entered, however, the ability to correct and award is typically limited to clarification or correction of clerical, typographical, technical, or computational errors. The rules also make clear that an arbitrator correcting an award is not empowered to redetermine the merits of claims already decided. As a result, a party that believes AI improperly influenced the merits of an award may have few options other than seeking vacatur under the Federal Arbitration Act or a state arbitration statute such as the Florida Arbitration Code.

The issues raised in LaPaglia v. Valve Corp. illustrate the issue. In that case, a claimant asked the United States District Court for the Southern District of California to vacate an AAA arbitration award entered in favor of Valve Corp. Among other arguments, the claimant alleged that the arbitrator relied on AI so extensively that he “outsourced his adjudicative role.” The claimant pointed to allegations that the arbitrator had discussed using ChatGPT to write an article, wanted to complete the award before a planned trip, and issued a 29-page award shortly after the post-hearing briefing concluded. The claimant also alleged that the award contained signs of AI generation and referenced facts that were not presented at the hearing or supported by the record.

The court did not reach the merits of the AI issue. Instead, it dismissed the petition on jurisdictional grounds after concluding that it lacked subject matter jurisdiction. That means the LaPaglia case did not resolve when an arbitrator's use of AI becomes improper, but it does show how these arguments may be framed in future challenges.

If the AAA’s AI Standards had applied and the claimant’s allegations had been proven, several requirements could have been implicated. The standards instruct neutrals to preserve human perspective and judgment, rely on direct review of evidence and reasoned deliberation, verify AI outputs against trusted materials, avoid shortcuts that compromise quality, and evaluate each dispute according to its unique context. Those requirements go to the core concern: AI may assist the decision-maker, but it should not replace the decision-maker.

The harder question is where to draw the line. An arbitrator who asks AI to review the record and generate a ruling with little or no independent analysis may be vulnerable to an argument that the award was the product of improper delegation. By contrast, an arbitrator who has already reached findings of fact and conclusions of law, and uses AI only as a drafting or organizational tool, may be on firmer ground even if the final award contains errors or imperfect reasoning.

For parties and counsel, the practical takeaway is not that AI must be avoided altogether. The better takeaway is that AI use in arbitration should be transparent, controlled, and anchored in human judgment. Parties may also want to consider addressing AI use in procedural orders, arbitrator disclosures, confidentiality protocols, and expectations for award preparation. As AI becomes more common in ADR, courts, arbitral institutions, and legislatures will likely be asked to provide clearer answers about proof, access to prompts or search histories, and the level of misuse required to disqualify an arbitrator or vacate an award.

The promise of AI in arbitration is real, but so are the risks. The next phase of the debate will not be whether AI can be used, but how to ensure it remains a tool for efficiency rather than a substitute for independent arbitral judgment.