Introduction
India’s arbitration ecosystem has spent the last decade trying to shed a reputation for delay and heavy court interference, with successive amendments to the Arbitration and Conciliation Act, 1996 and a stated national ambition to become a global arbitration hub by 2030. Into this reform push has arrived artificial intelligence now used by Indian law firms, arbitral institutions, and even constitutional courts for research, translation, and document review. AI offers a genuine answer to some of Indian arbitration’s oldest complaints: cost, delay, and an overburdened judicial ecosystem that arbitration was meant to relieve.
But arbitration’s value proposition in India rests on the same two pillars it does everywhere else: due process (encoded in Sections 18 and 34 of the Act) and confidentiality (now expressly codified in Section 42A). This article looks at how AI is being used in and around Indian arbitration, the specific due process and confidentiality risks it raises under Indian law, and where the regulatory and institutional response currently stands.
The Indian Statutory and Institutional Backdrop
Unlike some Western jurisdictions, India does not yet have a dedicated AI-in-arbitration guideline of its own. The Arbitration and Conciliation Act, 1996 modelled on the UNCITRAL Model Law is silent on AI use, and the Draft Arbitration and Conciliation (Amendment) Bill, 2024 does not squarely address it either, though commentators have argued it should. This places India somewhat behind jurisdictions such as Singapore, where SIAC’s 2025 Rules and related guidance already speak to AI use, and behind soft-law bodies like SVAMC and CIArb, whose frameworks Indian tribunals can voluntarily adopt but which have no domestic institutional counterpart yet.
That said, India is not starting from zero. Key elements of the existing framework are directly relevant to how AI should be used:
- Section 42A of the Act imposes a statutory confidentiality obligation on arbitrators, institutions, and parties, one of the more emphatic confidentiality provisions found in any Model Law jurisdiction, and one that any AI tool touching case data must be squared against.
- Section 34, governing setting aside of awards, includes grounds such as a party being unable to present its case and the award being in conflict with India’s public policy both of which become newly relevant if an award is later shown to rest on unverified or undisclosed AI-generated reasoning.
- Section 18 codifies the equal treatment of parties, which is implicated if one side’s access to sophisticated AI tools creates a material imbalance in the proceedings.
- The Digital Personal Data Protection Act, 2023 (DPDP Act) now overlays arbitration practice generally, requiring data minimisation, purpose limitation, and consent for processing personal data obligations squarely triggered when confidential witness statements, financial data, or personal information are fed into AI tools, especially cloud-based or foreign-hosted ones.
- Institutional rules of India’s leading arbitral bodies the Mumbai Centre for International Arbitration (MCIA), the Delhi International Arbitration Centre (DIAC), and the Indian Council of Arbitration (ICA)currently do not contain AI-specific provisions, leaving tribunals to fill the gap through procedural orders, much as their counterparts do internationally.
Where AI Is Already Being Used
Indian arbitration practitioners and law firms are, at a personal and firm level, already using generative AI for legal research, first-draft memoranda, translation of vernacular-language documents and evidence, and large-scale document review in complex commercial and construction disputes the same use cases seen internationally, but without a sector-specific Indian framework governing them.
In the judiciary, as a preview of what arbitral institutions may adopt. India’s own courts offer a useful, cautious template. The Supreme Court’s e-Committee has deployed SUVAS (Supreme Court Vidhik Anuvaad Software) for translating judgments into regional languages it translated over 31,000 of roughly 36,000 Supreme Court judgments by 2023 and SUPACE (Supreme Court Portal for Assistance in Court Efficiency), an AI research tool that extracts facts, legal provisions, and precedent for judges. Notably, SUPACE remains confined to an experimental, judge-supervised pilot, and the Court has been explicit that the tool is not meant to spill over into actual decision-making a distinction between AI-as-research-aid and AI-as-decision-maker that maps directly onto the non-delegation principle arbitral tribunals are expected to observe.
The Due Process Problem, Under Indian Law Specifically
If an arbitrator’s award is later shown to have relied on AI-generated legal analysis or fact finding without independent verification, a challenge under Section 34 becomes a live risk both on the ground that a party was unable to present its case (if it never had a chance to address AI-derived material outside the record) and on public-policy grounds, given Indian courts’ historically searching review of reasoned awards. The Indian judiciary’s own posture treating SUPACE strictly as a research aid, not a decision-making substitute is instructive for how tribunals should treat generative AI in drafting reasoning or awards.
Indian courts face a pendency of several crore cases nationwide, a backdrop that makes the efficiency promise of AI attractive but also raises the stakes of getting verification wrong at scale. Generative AI can sometimes produce fabricated citations or incorrect statements of Indian law. These errors may go unnoticed without careful verification, particularly because reliable and freely accessible Indian case law databases are less widely available than those in jurisdictions such as the US and UK.
In a jurisdiction where AI literacy and access to enterprise-grade legal AI tools vary significantly between large firms representing well-resourced corporate parties and smaller firms or self-represented parties in domestic arbitrations, the equal-treatment principle takes on a distinctly Indian dimension: tribunals may need to actively manage AI-driven disparities in the sophistication or volume of submissions, rather than assume rough parity as might be more reasonable in jurisdictions with more evenly distributed legal-tech access.
The Confidentiality Problem, Under Indian Law Specifically
India’s statutory confidentiality obligation is comparatively strong, but it was drafted without AI in mind. Inputting pleadings, witness statements, or financial data into a general purpose, cloud-hosted AI tool particularly one whose terms permit the provider to retain or review submitted data can plausibly constitute the kind of disclosure Section 42A is meant to prevent, even where no human at the counterparty ever sees the material.
Since witness statements and evidence in commercial arbitration frequently contain personal data (employee records, financial details, communications), the DPDP Act’s requirements , data minimisation, defined purpose, and consent apply in parallel with contractual confidentiality obligations. Counsel using AI tools for document review or drafting should treat DPDP compliance as a distinct checklist item, not an afterthought folded into general confidentiality practice.
Many widely used AI tools process data on servers outside India. Combined with the DPDP Act’s data-transfer provisions and sector-specific restrictions (for instance, in disputes involving regulated sectors like banking or telecom), this raises a compliance question that has no settled answer yet in Indian arbitration practice: does using a foreign-hosted AI tool on Indian arbitration data trigger transfer restrictions, and if so, whose obligation is it to check?
Where India Stands Relative to Its International Peers
Internationally, bodies such as SVAMC (2024) and CIArb (2025) have published dedicated AI guidelines, the SCC and VIAC have issued their own notes, and SIAC’s 2025 Rules already reference AI-related practice whereas India has no dedicated AI-in-arbitration guideline and the Draft Amendment Bill, 2024 does not address AI at all. India does, however, have some structural advantages: an express statutory confidentiality obligation under Section 42A (compared to the more contract- or rule based confidentiality found in many other jurisdictions), a dedicated data protection overlay in the DPDP Act, 2023 (paralleling the GDPR in the EU), and, in SUPACE’s judge-supervised, non-decisional design, a workable domestic template for how AI-as-decision-aid should function even as India’s own arbitral institutions, MCIA, DIAC, and ICA, have yet to issue rules addressing AI use directly.
This gap is not unique to India, but it is more consequential given the country’s stated ambitions in institutional arbitration a credible AI governance framework is likely to be one of the softer, less visible factors international parties weigh when choosing an Indian seat over Singapore, London, or Paris.
Practical Recommendations for Indian Practitioners and Tribunals
- Address AI use expressly in procedural orders, since neither the Act nor institutional rules currently do so tribunals seated in India should not assume silence means permission without conditions.
- Treat Section 42A as extending to AI tool selection. Before feeding case materials into any AI platform, confirm data handling and retention terms, and prefer tools with enterprise agreements over free consumer tools for anything touching pleadings or evidence.
- Build DPDP Act compliance into AI use protocols wherever witness statements, employee records, or other personal data are involved, separately from general confidentiality undertakings.
- Independently verify all AI-generated legal research against Indian primary sources case law, statutes, and SCC/AIR reporters given the documented risk of AI tools misstating or fabricating propositions of Indian law.
- Preserve non-delegation as a bright line, following the model the Supreme Court itself has adopted with SUPACE: AI may assist research, drafting, and translation, but the arbitrator’s own analysis of facts, law, and evidence must remain unmistakably the arbitrator’s.
- As MCIA, DIAC, and ICA come under pressure to match SIAC and international soft-law developments, practitioners should expect Indian institutional guidance to emerge in the near term and should be prepared to adapt.
AI’s efficiency case in Indian arbitration is compelling a system trying to escape decades of delay and court backlog has obvious reasons to embrace tools that speed up research, translation, and document review. But India’s arbitration framework carries its own specific due process guarantees under Sections 18 and 34, its own unusually explicit confidentiality mandate in Section 42A, and its own overlapping data protection regime in the DPDP Act all of which predate generative AI and none of which were designed with it in mind. Until Indian arbitral institutions or a future amendment to the Act catch up with the soft-law frameworks already operating internationally, the burden falls on tribunals and counsel to build AI governance into each case through procedural orders, informed by the same non-delegation principle the Supreme Court itself has applied to its own AI tools. Getting this right is not just a matter of avoiding a Section 34 challenge it is part of what will determine whether India’s arbitration ecosystem is trusted as a serious global seat of choice.