Shauryaveer Chaudhry is a third-year student at the National Law School of India University.

Introduction
The use of Artificial Intelligence(“AI”) has greatly improved the efficiency of businesses, and many scholars argue that the use of AI may help increase the speed of the arbitral process and therefore must be integrated within it.

This paper tries to understand the use of AI in the adjudicatory process of arbitration in India and eventually argues that the use of AI should be a ground to set aside an award under section 34 of the Arbitration and Conciliation Act, 1996 (“the Act”). India has no existing policy for the use of AI in arbitration. Therefore, this paper tries to understand how India should approach the AI question by drawing on jurisdictions that already have limited guidelines on how to govern AI. The central subject of this paper is the litigant.

I approach this article by firstly, drawing on John LaPaglia versus Valve Corporation, a case that directly dealt with this problem and gives a coherent policy structure for the same. Secondly, I identify the potential problems with using AI in the adjudicatory process. Thirdly, I situate this in the Indian context and argue that the use of AI in itself should be a ground to set aside an arbitral award under section 34 of the Act at the stage of adjudication by drawing on the principles of LaPaglia. Finally, I compare guidelines in force in other jurisdictions to find coherent policies that deal with AI.

While limited literature does exist on the use of AI in the legal field, nothing specific to the domain of arbitration existed in this regard, until the Delhi Arbitration Weekend 2025 held a seminar on it. Therefore, in this regard, a coherent policy for dealing with AI in the field of arbitration is needed. This is what this article attempts to achieve by building on the discussions held at this seminar.

LaPaglia
The question as to the use of AI while drafting an arbitral award has already been under consideration by courts around the globe. One example of this is the case of John LaPaglia versus Valve Corporation before the Southern District of California, where compensation for higher prices paid as a result of Valve’s antitrust violations was demanded. The matter subsequently went for arbitration, and the arbitral award was challenged under the Federal Arbitration Act. The third submission in this case was that the Arbitrator had outsourced his adjudicative role to Artificial Intelligence.

The claim that the arbitrator had used AI was substantiated in 3 supplementary manners:

  1. During breaks in the arbitration, the arbitrator revealed that he wanted to conclude the arbitration with haste since he had a trip scheduled to the Galapagos Islands.
  2. The arbitrator further revealed that he uses AI to write articles for him.
  3. The award bore hallmarks of AI drafting.

Based on further investigation by the court, it was held that AI was used to draft the award, supplanting the arbitrator’s fact-finding and adjudicative role with facts found by a machine. Several errors were found in the award thus drafted, which included the citation of facts that were untrue or not presented in the record.

The court eventually vacated this award. The court reiterated that the arbitrator is responsible for supplying a written decision which is supplemented by a statement of reasons for its holding. The Arbitrator’s reliance on generative AI to replace its own role and the parties’ submissions betrays the parties’ expectation of a well-reasoned decision rendered by a human arbitrator. The court further opined that just as awards have been vacated when the decision-making is outsourced to a person other than the arbitrator so appointed, the court must also vacate those awards wherein decision-making is outsourced to Artificial Intelligence.

This reasoning of the court could be seen to reflect public policy that strikes to the very root of the judicial or quasi-judicial process. The focal point of any adjudicatory process at the end of the day must be the litigant, and arbitration focuses on the intent of the parties. They are the aggrieved parties who approach any forum when injustice is carried out against them and therefore must be placed at the center stage when such changes in the adjudicatory process are envisioned. Therefore, the question to be asked here is whether the litigants would trust a machine to adjudicate their disputes. I argue that the answer to this question would be in the negative. The human arbitrator has the ability to use his human mind in a judicial manner while employing principles of equity, empathy, and the skill of reading between the lines. These are qualities that are lacking in AI.

Problems with using AI in the Arbitral Process
There exist three dominant problems with allowing Artificial Intelligence to be involved in the core functions of the arbitrator for the purposes of fact finding and decision making:

Firstly, there lie issues with confidentiality. At the end of the day, AI software is operated by private corporations and is usually open-sourced. The confidentiality concern arises particularly where parties use proprietary, cloud-based AI tools operated by private corporations, since information entered into such systems may be processed or retained by third parties beyond the parties’ control.   Current frameworks do not presume liability for a breach of data caused by artificial intelligence. Moreover, sooner or later AI will no longer be considered a simple tool in the hands of the owner, the user, or the manufacturer.

This is further complicated by the fact that arbitral proceedings are meant to be confidential. The primary reason as to why parties choose the avenue of arbitration is the confidential nature of the proceedings. Parties to the proceedings include large commercial entities with privileged information and do not wish for information in the proceedings to be released to the public at large since it may affect the business of such parties. The same was recognized by a High- Level Committee of the Ministry of Law and Justice of India.

Secondly, in line with the decision taken in LaPaglia, AI is not fit for adjudication due to an inability to use equity, empathy, and the skill of reading between the lines. The New York Convention explicitly states that there is no scope for judicial review in the process of fact-finding or evidence collection in arbitral disputes. The same is entrenched in Indian Arbitration jurisprudence. This can be attributed to the trust of the institution in the arbitration tribunal. This trust will be undermined if AI is used in such a process and will open the floodgates for courts to review basic findings of fact, thereby increasing the court’s burden and defeating the very purpose of arbitration.

Thirdly, there exist problems with the technology itself. Generative AIs tend to hallucinate cases with nonexistent parties and facts. Their interpretation of real cases is based on either false or weak equivalence. Perhaps the most unsettling of these problems is the issue of algorithmic biases. Studies have shown that AIs are likely to be biased against people of colour and gender-based minorities.  This is extremely problematic and will inevitably set the stage for institutionalised discrimination. The existence of biases within AI is testament to the fact that Artificial Intelligence must be kept away from adjudicatory functions of tribunals.

An Exception to Section 34?
Now that the perils of using Artificial Intelligence in the adjudicatory functions of the arbitrator have been established, another important question must be asked. Whether the use of generative AI in itself should be a ground to merit the setting aside of an award under section 34 of the Act, or whether it must meet one of the explicit exceptions already given in the section to set aside an arbitral award.

Indian courts have not yet settled this position of law, so it may be worth looking at other jurisdictions for an answer to this question. The case of LaPaglia highlighted the fundamental errors that AI made while reaching a decision, such as ignoring the submissions of the parties, citing wrong facts, and considering issues peripheral to the main question at hand. However, a careful reading of the judgement would classify this as obiter, and the main conclusion of the court relies on the fact that an arbitrator’s reliance on generative AI undermines the requirement for a human-rendered, well-reasoned decision, which is akin to outsourcing adjudication to an unappointed third party.

The same reasoning, however, cannot be automatically transplanted to India. The threshold for maintaining a challenge under Section 34 is higher than that. Courts may choose to interpret it in a stricter manner and only set aside awards when the AI has violated principles of natural justice, ignored the submissions made by the parties, incorrectly interprets and applies precedents, or on grounds of patent illegality.

I, however, argue that without the further development of these Artificial Intelligence technologies, the approach taken by the Southern California District Court may be a better recourse for courts in India to take. This approach would place the litigants at the center of the adjudicatory process until we reach a stage where AI can be integrated for smaller roles in the arbitration process. What this implies is that any core aspect of the process of litigation, when delegated to Artificial Intelligence, should be considered as a ground to challenge and set aside an award under Section 34. This ground, however, should be incorporated notwithstanding party autonomy. If the parties to a contract explicitly agree to allow AI to be incorporated in their arbitration process, such awards should be allowed to be challenged under Section 34. 

Looking Forward
With the increase in permeability of AI technologies in real life, one cannot help but wonder that we will reach a stage where some of the problems that AI presents at the moment may be resolved. In such a scenario, the call for integrating AI in the arbitration process (for procedural aspects at least) may become stronger within the legal fraternity. Therefore, it is imperative to understand what other jurisdictions are doing to frame guidelines for the use of AI in the arbitration process, and to decide the best practices that should be adopted in India.

The French Code of Civil Procedure has, in a manner, codified the LaPaglia line of reasoning. Article 1450 of the Code expressly states, “An arbitrator may only be a natural person enjoying the full exercise of his or her rights.” This ensures that AI cannot take over the adjudicatory role of the arbitrator.

New South Wales, going in a different direction, provides yet another set of interesting guidelines for the use of AI. They permit the use of AI in the preparation of submissions and arguments by lawyers, provided that the counsel manually recheck the accuracy of the contents and references that AI has made. By assigning the burden on the parties for the use of AI, the core adjudicatory function of the arbitrator remains intact. Another principle entrenched here is the concept of full disclosure of the use of AI.

The guidelines of the European Union provide for the judge to be personally liable for what is published in their name. Thus, any error made by the AI would be considered to be an error of the judge itself.

The question of confidentiality in the use of AI was tackled at length in the Delhi Arbitration Weekend 2025. A two-prong safeguard was suggested for this:

  1. Selection and design of the relevant AI tool – This has to ensure a model that is not open source but rather at an enterprise-level subscription;
  2. Control exercised by the tribunal over the use – This is to ensure the tool respects the confidential functions and limits of the AI tool.

While these guidelines may be a useful tool somewhere in the future, as it stands in India, AI has not been integrated into the arbitral process and must remain outside the domain of the core areas that represent arbitration until the technology is developed and substantially improved. The arbitration process must remain neutral and preserve the due process of law, even if some amount of speed is compromised in the process.