[This guest post is authored by Shubham Shrivastva, Penultimate Year Law Student, Dharmashastra National Law University]

Consider the following thought exercise. Into an international arbitration, two legal teams go. At stake are the multi-billion-dollar investments of a western energy firm pitted against a state-owned Asian enterprise. Both sides use advanced document review platforms equipped with artificial intelligence (AI). The claimant uses AI based on a large language model from Silicon Valley while the respondent uses one from Shanghai. The 400.000-page evidentiary record is identical in both tools. Both models provide summaries, highlight key precedents, and help counsel piece together factual timelines but present contradictory factual conclusion. 

Not because the lawyers are unscrupulous. Not because the evidence is ambiguous. Since the two AI systems are trained on data constructed under distinct geopolitical conditions differing legal systems, state narratives and understanding of “market practice,” “sovereign immunity,” or “force majeure”. The two narratives diverge in their legal interpretations.  No disclose is made. The tribunal never knows. An award is rendered. This is not a theoretical warning about some distant future. This is a description of an arbitral environment that already exists. One which international dispute resolution scholarship has largely overlooked. 

The Fracture Nobody Is Talking About

Generally, international legal scholars have expended a significant amount of effort in discussion and debating the question of whether AI can be an arbitrator, how AI-based decision-support tools can be regulated and what disclosure obligations counsel should have. These are sound issues to consider. 

The term “splinternet” (splintering of the global internet into separate geopolitical internet realms) has been discussed thoroughly in the technology policy literature. This splintering is less documented but has become part of the procedural anatomy of international arbitration. The world’s top arbitral institutions operate today on digital platforms, on cloud infrastructure and on digital software which are located on either side of the digital divide, aligned with the United States or China. The international chamber of commerce runs on a western digital system. The China international economic and trade arbitration commission is pat of the other. 

This is not merely an administrative matter, but an architectural one. These ecosystems embed epistemologies when AI tools for e-Discovery, legal research, translation, evidence analysis, and, more recently, decision support are used. They are not just ‘fact processors. They indicate whose facts are counted. 

Weaponized Interdependence Enters the Hearing Room

International relations scholars Henry Farrell and Abraham Newman coined “weaponized interdependence” to describe how a state’s structural dominance in the global economy-as a financial informational, and logistical hub- can be deployedas instrument of coercion. The next frontier of that leverage and arbitration is particularly exposed to it is the AI infrastructure that supports international legal practice. What is a state imposes technology sanctions which limit a foreign party’s ability to utilize wester technology for e0discovery while the arbitration is ongoing? Or when one of the experts appointed by tribunal is unable to conduct a forensic analysis on financial records because of the laws of country or province I which the records are stored. Or when the case management system provided by an arbitral institution relies on artificial intelligence and the confidentiality risks are posed by access to foreign intelligence laws and rules that are not currently included in the arbitral institution’s case management system.

This is not a ‘special case’. They are the predictable results of installing into arbitration’s most sensitive processes on infrastructure on which powerful states have proven themselves willing to play the game. This leaves the arbitrator with structural due process problems, not related to his actions, but to the technology layer under him. 

Epistemic Sovereignty: A New Concept For A New Crisis

Naming this problem clearly requires mew vocabulary. The term “epistemic sovereignty” is a borrowing and adaption from science and technology studies, and refers to the ability of a geopolitical player to determine factual truths by controlling knowledge-producing infrastructure. The implication in arbitration is a troubling one: the fact that an AI system’s geopolitical allegiance determines its factual universe means that arbitration is epistemically fragmented AI environments are, in a meaningful sense, not neutral. 

This extends beyond the normal meaning of bias. The issue isn’t just that an AI tool that has been trained largely on US law will devalue Chinese legal principles, or that an AI model trained on Chinese commercial data will overvalue state-enterprise relations, which are viewed with suspicion in western legal systems. The deeper issue is one of structural invisibility.  An AI system’s epistemics are generally opaque, proprietary and not subject to a current challenge procedure. The arbitral record, however, can be the silent record created by geopolitical epistemics not discernible or interrogable by any party, counsel, or the tribunal. 

Why Current Procedural Frameworks Cannot Cope

The New York Convention, enacted in 1958, deals with recognition and enforcement of arbitral awards. The UNCITRALModel Law is limited to the procedural aspects of fairness in a largely pre-digital context. Recent years have seen guidance from ICC, SIAC, LCIA and HKIAC on cybersecurity. And to a similar degree guidance on disclosing AI usage, though none of these measures is systematic mandatory, or addressed to systemic infrastructure risk.

None of these frameworks have anything like a disclosure obligation for “AI provenance”. Such an obligation would require parties to disclose which AI systems they used, on which platforms, with which data sources, etc. None present any evidence of AI epistemic bias that would justify challenging the award. None demand geopolitical Neutrality Audits of the technology power behind arbitral institutions. The divide is more than just a technical one. It embodies an underlying premise that international arbitration is a neutral space, where justice itself is neutral and where it is above geopolitics. It was always an aspiration rather than a guaranteed reality.

What Reform Must Look Like

Three levels of thinking are needed to move forward. 

In terms of procedure, disclosure of AI usage should be made mandatory in international arbitration proceedings. Firms that use AI In document review, evidence analysis, or legal research should be required to provide information on the tool’s developer, training data provenance and where the tool is hosted. Is the 

At the institutional level, leading arbitral institutions should order independent geopolitical neutrality audits of technologies that are used. The question of whether an institution’s case management system, AI-based translationservices or handballing of documents are based on infrastructure that can be accessed by foreign states is not a regular IT question, but should be treated as a fundamental question of institutional integrity. 

From doctrinal point of view, arbitration law requires a new jurisprudential category: the challenge of an award based on undisclosed algorithmic bias of geopolitical origin. While this is admittedly a high bar set, its articulation matters: it coveys to institutions, parties, and counsel that AI epistemics are not merely a theoretical issue but one that is live and present in ongoing proceedings. 

The Stakes Are Not Abstract

The legitimacy of international arbitration rest on one overarching assurance: that there is a supranational, neutral forum for dispute resolution where states and parties can solve their disputes fairly. It is the promise which induces parties to accept its jurisdiction and to provide finding for its institutions and enforcement of its awards. 

If the AI that drives arbitration is geopolitically divided, and if the tool that build the evidentiary record, aid counsel and support institutional administration are epistemically aligned to competing geopolitical blocs, then the neutrality of arbitration is not only under threat at its edges. It’s cracked at the core.  The issue is not whether international arbitration can overcome the geopolitics. It has always been in geopolitical tension. The question then becomes, can it withstand the invisible geopolitics the ones that do not enter through the arbitrator’s nationalities or the institution’s home jurisdiction, but rather through the epistemics of the algorithm running in the background? This is a question that needs to be answered now-before the net award is made, not after.

***