--- title: "The Fourth Party at the Indian Tribunal Table" date: 2026-05-25 author: "Navod Prasannan" url: https://ksandk.com/litigation/ai-arbitration-india-fourth-party-tribunal/ --- # The Fourth Party at the Indian Tribunal Table Posted On - 25 May, 2026 • By - Navod Prasannan ![The Fourth Party at the Indian Tribunal Table](https://ksandk.com/wp-content/uploads/The-Fourth-Party-at-the-Indian-Tribunal-Table.webp) When two parties agree to arbitrate, they agree to place their dispute before a person, or a panel of persons, whom they trust to decide it. That is the whole of the bargain. Everything else, the seat, the rules, the language and the timetable, is machinery built around a single human act of judgment. It is worth holding that picture in mind, because a quiet change is taking place at the tribunal table. A new presence has pulled up a chair, and it is neither party, nor counsel, nor the arbitrator. Scholars of online dispute resolution gave it a name some years ago. Drawing on the work of Ethan Katsh and Janet Rifkin, they called technology the fourth party, sitting alongside the two disputants and the neutral and increasingly shaping what happens between them. The phrase was coined for the modest software of an earlier internet. It fits the arbitration room of 2026 far better than it fitted the one it was written for. The argument of this piece is narrow and important. So long as artificial intelligence in arbitration does the work of a clerk, it raises little that the existing law cannot handle. The moment it begins to shape the decision rather than merely speed up the typing, it stops behaving like a tool and starts behaving like a participant, and at that moment a set of Indian rules built entirely around human actors begins to misfire. The fourth party is not a metaphor to be admired. It is a problem to be located precisely, and Indian arbitration law, as it happens, gives us unusually sharp instruments for locating it. ## **What the fourth party actually means** The fourth party idea is best understood by contrast with the third. The third party in any dispute is the neutral, the mediator or the arbitrator, brought in to do what the two sides cannot do for themselves. The fourth party is the technology that increasingly assists, and sometimes supplants, that neutral. In its original and innocent form it was the platform that scheduled the mediation and held the documents. In its present form it is a system that can read the evidence, draft the analysis and propose the outcome. The point of the phrase is to make us notice that the technology is no longer merely infrastructure sitting in the background. It has moved into the foreground, close enough to the decision to deserve a name. It is essential to separate two modes of deployment of AI, because the entire analysis turns on the distinction. The first is clerical. A tool that transcribes a hearing, translates a document, organises an index or corrects the spelling in a draft does work that is real but not dispositive, and nobody sensible loses sleep over it. The second is dispositive, or close to it. A tool that weighs the evidence, assesses credibility, decides which line of authority to prefer or drafts the operative reasoning of an award is doing the very thing the parties appointed a human to do. The fourth party becomes a legal problem only in this second mode, and much of the confusion in the current debate comes from a failure to say which mode is in issue. ## **The arbitrator we choose, and why we choose that one** Indian law, like the law of most arbitral jurisdictions, treats the appointment of an arbitrator as personal. The choice is intuitu personae, made in respect of the particular individual and their particular qualities, their expertise, their judgment and their reputation for fairness. This is not sentiment. It is the reason the parties are bound by the award of a person whom they did not have to accept and could have rejected. Because the appointment is personal, the mandate that flows from it is non delegable. An arbitrator may take administrative help, but may not hand over the decision to someone else, because the someone else is not the person the parties chose. This principle already has a well known stress point that long predates artificial intelligence, namely the tribunal secretary. The international debate about how much a tribunal secretary may properly do, and at what point assistance shades into the secretary becoming a fourth arbitrator who improperly shares in the decision, is precisely the debate we are about to have again, only this time with a machine in the secretary’s chair. The lesson from that earlier debate applies directly. The line was never drawn at research, summarising or drafting, all of which a secretary may properly do. It was drawn at the decision itself, which must remain the arbitrator’s own. A language model is, for this purpose, a tireless and untrustworthy tribunal secretary, and the same line governs it. It may assist up to the point of decision. It may not make the decision, and it must not be permitted to make the decision in substance while a human merely ratifies the output. The Indian reality sharpens this point rather than softening it. A very large share of Indian arbitration is conducted by sole arbitrators, frequently retired judges, who carry heavy lists and lean, quite properly, on juniors and clerks to manage the paper. The institutional scrutiny that a tribunal secretary attracts in a large international reference is often simply absent in a domestic one. Into that environment now arrives a tool that will draft a confident analysis on request, at no marginal cost, at any hour of the night. The temptation to let it do more than it should is therefore greatest exactly where the supervision is thinnest. A rule that depends on busy sole arbitrators policing themselves, with no institutional check standing behind them, is a rule that will be honoured unevenly, and that is a reason to make the expectation explicit rather than to leave it to good intentions. ## **A disclosure regime aimed at the wrong actor** Here the fourth party exposes a genuine gap. Indian law polices the integrity of the decision maker through Section 12 of the Arbitration and Conciliation Act, 1996, read with the Fifth and Seventh Schedules introduced in 2015. An arbitrator must disclose any circumstances likely to give rise to justifiable doubts about independence or impartiality, and certain relationships render a person ineligible altogether. The entire apparatus is trained on the human arbitrator. It asks whether that person is independent, whether that person has a conflict, and whether that person can be trusted to hold the balance even. It asks nothing at all about the tool that may be shaping that person’s reasoning, because when the apparatus was designed there was no such tool. Consider how odd this sits. If a retired judge sitting as a sole arbitrator has a remote past association with one of the parties, the law requires disclosure and may require recusal. If the same arbitrator runs the entire dispute through a commercial artificial intelligence system whose training data, commercial alignments and systematic leanings are entirely unknown, the law currently requires nothing, because the system is treated as a pencil rather than as a participant. The disclosure regime is looking hard at the arbitrator and not at all at the fourth party that may be doing a meaningful part of the arbitrator’s thinking. That is no criticism of the draftsman of 2015, who could not have foreseen it. It is simply an identification of where the next reform must look. ## **Equality of arms, and a recent warning from the Supreme Court** The Indian courts have, if anything, been moving in the opposite direction to complacency about who controls the decision maker. In *Central Organisation for Railway Electrification v. ECI-SPIC-SMO-MCML (JV)* reported as 2024 INSC 857, a Constitution Bench of the Supreme Court held in November 2024 that clauses allowing one party to unilaterally appoint a sole arbitrator, or to confine the other side’s choice to a panel curated by that one party, offend the principle of equality under the Act and the guarantee of equality before the law under Article 14 of the Constitution. The reasoning built on earlier decisions such as *TRF Ltd. v. Energo Engineering Projects Ltd.* reported as (2017) 8 SCC 377 and *Perkins Eastman Architects DPC v. HSCC (India) Ltd.* reported as (2020) 20 SCC 760, which established that a person ineligible to act as an arbitrator cannot validly appoint one either. The thread running through these decisions is a deep judicial insistence that neither side should hold a structural advantage in constituting the tribunal, because the tribunal is the heart of the bargain and must be equally the tribunal of both parties. Now place the fourth party against that thread. If the arbitral process leans heavily on an artificial intelligence system, and one party has access to a more powerful system than the other, or worse, if the tribunal itself relies on a system supplied or shaped by interests aligned with one side, the equality that the Supreme Court was so anxious to protect is quietly disturbed, not at the visible stage of appointment but at the invisible stage of reasoning. The Court closed the front door to a stacked tribunal. The fourth party can walk in through a window the Court was not yet looking at. Section 18 of the Act states the same value in the language of procedure, requiring that the parties be treated with equality and that each be given a full opportunity to present its case. A fourth party that one side can afford and the other cannot, or that systematically favours the kind of argument it was trained on, is a Section 18 problem wearing technological clothing. The provision is broad enough to reach it. What is missing, for now, is the habit of looking. The equality concern is not an abstract one either. Indian arbitration frequently pits a well resourced entity, a public sector undertaking, a bank or a large contractor, against an individual, a small supplier or a sub-contractor. The Supreme Court in the railway electrification case was alive to exactly this imbalance when it struck down panels curated by the stronger party. Now imagine that the stronger party deploys a sophisticated and expensive analytical system across the entire reference while the weaker party cannot, or that the tribunal itself adopts a tool whose defaults quietly reward the kind of voluminous, well organised submission that only the stronger party can produce. The inequality does not announce itself. It is laundered through the appearance of neutral efficiency. The lesson of the railway electrification judgment is that Indian law cares about structural advantage in the constitution and the conduct of the tribunal, and the fourth party is entirely capable of delivering such an advantage through the back door. ## **The honest counter** The strongest objection to all of this must be met head on, because it is a good one. Every arbitrator already uses tools. They use Manupatra and SCC Online, they use juniors and clerks, they use their own libraries and their own past awards. Nobody calls a legal database a fourth party or demands that it be disclosed and conflict checked. Why should a language model be any different? The answer is that the difference lies not in the tool being electronic but in the tool generating rather than retrieving. A database returns what a human asked for, and the human then evaluates it. A generative system proposes conclusions, frames the analysis and supplies reasoning that a tired or busy human may adopt with less scrutiny than they would give a junior, precisely because the output reads as finished and confident. The risk is not the technology in itself. It is the seductive completeness of the output and the human tendency to defer to it. That is what makes the generative tool a candidate for participant status when the database never was. This is not armchair psychology. The tendency to over rely on automated output, sometimes called automation bias, is well documented, and it is strongest under precisely the conditions in which arbitrators work, namely time pressure, large volumes of material and a confident, fluent answer that arrives already formatted. A junior who hands up a weak note invites correction, because the human relationship makes scrutiny natural. A machine that hands up a polished one invites adoption, because there is no relationship to mediate the scrutiny and the polish itself does the persuading. The danger is therefore not that arbitrators are careless. It is that the tool is built to be believed, and any rule that ignores this human factor is regulating the wrong thing. There is a fair reply to my own argument too, and intellectual honesty requires me to state it. One might say that all of this collapses into the simple and existing rule that the arbitrator must apply their own mind, so that no new concept of a fourth party is needed at all. I have some sympathy with that view, and in a sense the fourth party is only a vivid way of naming a failure of the old duty. But the vividness earns its keep. Naming the fourth party forces the system to ask a question it would otherwise skip, which is not merely whether the arbitrator applied their mind, but to what, and shaped by what, that mind was applied. The old duty asks about the arbitrator. The fourth party asks about the influence upon the arbitrator, and that is the question the present moment requires. ## **Where the line falls** So where does the line fall in practice? It falls; at influence over the dispositive reasoning. Below that line, in the clerical and the merely assistive, the fourth party is a tool, needs no special treatment, and the existing law is sufficient. At or above that line, where the system shapes the assessment of evidence, the choice of authority or the operative reasoning of the award, three things should follow. The arbitrator should disclose the use, because the parties are entitled to know what is in the room. The arbitrator must independently own every step of the reasoning, so that the award is genuinely their own and not a ratified output. And the institutions, when they next revise their rules, should extend the logic of Section 12 and the spirit of the railway electrification judgment to ask not only whether the human decision maker is independent, but whether the fourth party at the table is independent too. None of this requires heroic drafting. An institution could achieve most of it with a single default provision: that any use of a generative artificial intelligence tool by the tribunal which bears on the assessment of evidence or the reasoning of the award must be disclosed to the parties, that the tribunal remains personally responsible for every finding, and that any tool so used must meet stated standards of confidentiality and security. Parties who wished to contract out of the default could do so, by agreement, as party autonomy permits. Parties who said nothing would receive a sensible rule rather than a vacuum. That is precisely how arbitral institutions have introduced every other procedural innovation of the last two decades, from emergency arbitrators to expedited timelines, and there is no good reason the fourth party should be handled any differently. The fourth party is not a reason to keep artificial intelligence out of arbitration. It is a reason to keep it in its seat. Indian law has spent the last decade being unusually careful about who gets to constitute and influence a tribunal, and that care is exactly the right instinct to bring to this question. The arbitrator whom the parties chose must remain the one who decides. The fourth party may pull up a chair, take notes, hand up a draft and make itself useful in a hundred ways. What it may never do is pick up the pen that the parties placed, deliberately and personally, in a human hand. *Navod Prasannan, Advocate and Partner, Dispute Resolution, King Stubb & Kasiva, Chennai* --- ## Office Locations - [New Delhi](https://ksandk.com/locations/top-corporate-law-firm-in-delhi/) (HQ): +91-11-41318190 | info@ksandk.com - [Mumbai](https://ksandk.com/locations/top-corporate-law-firm-in-mumbai/): 3 offices (Nariman Point, Lower Parel, Andheri) | mumbai@ksandk.com - [Bangalore](https://ksandk.com/locations/top-corporate-law-firm-in-bangalore/): bangalore@ksandk.com - [Chennai](https://ksandk.com/locations/chennai/): chennai@ksandk.com - [Hyderabad](https://ksandk.com/locations/hyderabad/): hyderabad@ksandk.com - [Pune](https://ksandk.com/locations/pune/): pune@ksandk.com - [Kochi](https://ksandk.com/locations/kochi/): kochi@ksandk.com ## Contact - [Contact Page](https://ksandk.com/contact-us/) - General: info@ksandk.com | +91-11-41318190 - WhatsApp: +91-7428567444 - [Privacy Statement](https://ksandk.com/privacy-statement/) - [Terms of Use](https://ksandk.com/terms-of-use/)