Author : Sameer Panda from National Law University Odisha
ABSTRACT
The integration of Artificial Intelligence (AI) to quantify complex cross-border tax indemnities is a commercial necessity that introduces a profound evidentiary dilemma in international arbitration. While the Supreme Court of India’s Vidya Drolia framework affirms the arbitrability of private contractual tax indemnities, utilizing opaque “black box” machine learning models creates severe procedural friction. Specifically, the strict digital authentication mandates of Section 65B of the Indian Evidence Act conflict with the procedural flexibility granted by Section 19 of the Arbitration Act. Admitting unexplainable AI audits strips opposing counsel of their right to effective cross-examination, exposing arbitral awards to annulment under Section 34 for violating the fundamental policy of Indian law. Moving from theory to practice, the paper proposes a five-point ‘Practitioner’s Blueprint’ encompassing Algorithmic Consent and Transparency (ACT) clauses, confidentiality rings, Tandem Expert sponsorship, and Tribunal-Appointed Technical Experts (TATE). This structured approach enables Indian tribunals to harness algorithmic efficiency while safeguarding the structural
integrity and due process of the arbitral regime.
KEYWORDS:
International Commercial Arbitration, Algorithmic Proof, Tax Indemnities,
Section 65B, Due Process
I. INTRODUCTION
The landscape of international commercial arbitration is undergoing a seismic digital
transformation. As cross-border transactions become increasingly labyrinthine, the commercial disputes arising from them particularly those involving complex taxation warranties, transfer pricing adjustments, and contractual tax indemnities demand the reconciliation of colossal, global datasets. Traditionally, auditing these cross-border ledgers required immense human capital and protracted timelines. Today, multinational corporations and premier accounting firms are aggressively deploying Artificial Intelligence (“AI”) and machine learning algorithms to identify financial anomalies, determine arm’s length pricing, and calculate commercial damages at unprecedented speeds.1 However, the transition from the human forensic accountant to the “AI Auditor”
introduces a precarious challenge for arbitral tribunals: the evidentiary minefield of algorithmic opacity. Unlike traditional rule-based software, advanced machine learning models construct their own analytical pathways often referred to as the “black box” phenomenon making it exceptionally difficult to trace how the algorithm arrived at a specific financial conclusion.2In the context of arbitration, where the legitimacy of the award is anchored in the parties’ equal right to present their case and cross-examine evidence, relying on an inscrutable AI model threatens the very bedrock of procedural fairness.3 This tension is particularly acute in arbitrations seated in India. The Indian arbitral framework operates on a dichotomy regarding the admissibility of evidence. While Section 19 of the Arbitration and Conciliation Act, 1996 (the “Arbitration Act”) explicitly emancipates tribunals from the strictures of the Indian Evidence Act, 1872 (the “Evidence Act”),4 Indian jurisprudence heavily regulates the admission of electronic records to prevent data manipulation and algorithmic hallucinations.5 If a tribunal admits an AI-generated tax audit without requiring the party to disclose the underlying training data or algorithmic logic,
opposing counsel is entirely stripped of their right to effective cross-examination. Such a procedural defect invites acute judicial scrutiny under Section 34 of the Arbitration Act, risking the annulment of the award for contravening the “fundamental policy of Indian law.”6
II. THE JURISDICTIONAL THRESHOLD: ARBITRABILITY OF TAX INDEMNITIES POST-VIDYA DROLIA
Before evaluating the evidentiary admissibility of an “AI Auditor,” it is imperative to
establish the jurisdictional competence of an arbitral tribunal to adjudicate taxation-related claims. In India, the arbitrability of tax disputes presents a complex jurisprudential chasm. Taxation is fundamentally a sovereign prerogative; therefore, a careful delineation must be drawn between statutory tax liabilities owed to the State and contractual tax allocations
negotiated between private commercial entities.7
A. THE SOVEREIGN EXCEPTION AND THE STATUTORY BAR
Under Indian law, direct disputes contesting the validity, quantum, or applicability of a tax levied by the sovereign are strictly non-arbitrable. This prohibition is statutorily enshrined in Section 293 of the Income Tax Act, 1961, which expressly bars civil courts and by extension, arbitral tribunals from modifying or setting aside proceedings taken or orders made under the Act. 8The Indian judiciary has consistently held that where civil court jurisdiction is ousted in favor of specialized statutory tribunals (such as the Income Tax Appellate Tribunal), arbitral jurisdiction is concurrently ousted.9
This absolute bar was comprehensively detailed in the expert opinion of Sudipto Sarkar SA in the prominent Cairn Energy investment treaty arbitration, which clarified that the Indian hierarchy of legal norms explicitly prohibits private dispute resolution mechanisms from usurping the State’s taxation authority.10 Consequently, an arbitral tribunal seated in India lacks subject-matter jurisdiction (ratione materiae) to rule on direct tax evasions or statutory levies.
B. THE VIDYA DROLIA PARADIGM AND RIGHTS IN PERSONAM
To overcome this jurisdictional hurdle, practitioners must frame complex taxation claims
not as challenges to sovereign authority, but as private contractual indemnities. This distinction is governed by the landmark Supreme Court of India decision in Vidya Drolia v. Durga Trading Corporation (2021).11Expanding upon the foundational principles laid down in Booz Allen and Hamilton Inc., the Vidya Drolia bench established a definitive four-fold test to determine non-arbitrability.12
Under the Vidya Drolia framework, a dispute is non-arbitrable if it relates to actions in
rem (rights against the world at large), affects third-party rights, relates to inalienable sovereign functions, or is expressly non-arbitrable by statute.13 While a direct tax levy is an action in rem and a sovereign function, a contractual tax indemnity such as a warranty executed during a cross-border merger and acquisition (M&A) creates purely in personam rights (rights against a specific person or entity).14
Professor William W. Park’s tripartite division of fiscal disputes provides a vital theoretical construct here.15 Professor Park accurately categorizes these as “commercial transactions” disputes where arbitrators address tax matters that are purely incidental to basic contract claims between private parties. If Company A acquires Company B and later discovers massive
undisclosed transfer pricing liabilities, Company A’s claim for compensation under the Share Purchase Agreement is an arbitrable commercial dispute, not a sovereign tax dispute.
C. THE CONTRACTUAL REALITY AND THE IMPETUS FOR THE AI AUDITOR
It is within this carved-out space of in personam commercial indemnities that the true
complexity of modern arbitration lies. When tribunals adjudicate these cross-border tax
indemnities, they are routinely forced to unravel intricate, multi-layered corporate structures to determine actual liability, akin to navigating the “control conundrum” and piercing the corporate veil under India’s Significant Beneficial Ownership (SBO) regime.16 Resolving these indemnities requires parsing through dense, multi-jurisdictional financial webs and transfer pricing ledgers spanning years or even decades. The sheer volume of data required to quantify these contractual damages renders the traditional human forensic accountant highly inefficient, if not entirely obsolete. Thus, corporate parties are increasingly deploying AI Auditors to pierce these financial veils, identify hidden beneficial ownerships, and calculate precise indemnity figures.17 Having established that arbitral tribunals possess the requisite jurisdiction to hear these commercial tax indemnity claims, the inquiry must logically progress to the procedural viability of the evidence presented. If an AI Auditor is deployed to quantify an arbitrable claim, the tribunal must determine how to admit, evaluate, and test this algorithmic proof without fracturing the procedural integrity of the arbitration.
III. NAVIGATING THE EVIDENTIARY MINEFIELD: THE “BLACK BOX”
PROBLEM, SECTION 19 FLEXIBILITY, AND SECTION 65B CONSTRAINTS
Having established that complex commercial tax indemnities fall squarely within the
jurisdictional purview of arbitral tribunals, the inquiry must shift to the procedural mechanics of pleading and proving these claims. When a corporate claimant relies on an “AI Auditor” to analyze a multi-jurisdictional financial ledger and quantify a transfer pricing discrepancy, it introduces an epistemological shift in how evidence is generated and evaluated.18 This shift 16 creates a severe evidentiary minefield, primarily characterized by the algorithmic “black box” problem and a direct clash with Indian evidentiary statutes.19
A. THE “BLACK BOX” PROBLEM AND THE DEMISE OF TRADITIONAL CROSS- EXAMINATION
The fundamental utility of modern machine learning algorithms lies in their ability to
transcend traditional, rule-based programming. Unlike legacy software that follows linear, human-coded “if/then” instructions, advanced AI models particularly deep neural networks ingest massive datasets and independently construct their own inferential pathways to arrive at a conclusion.20 While this allows the AI Auditor to calculate complex tax liabilities with unprecedented speed, it results in profound algorithmic opacity, colloquially termed the “black box” problem.21
In the context of arbitral proof, this opacity is highly problematic. Under Section 26 of the Arbitration Act, parties possess a statutory right to put questions to an expert witness and present their own expert testimony to challenge the findings.22 Historically, if a human forensic accountant presented a transfer pricing audit, opposing counsel could effectively cross- examine them to expose methodological flaws, biased assumptions, or mathematical errors. However, one cannot cross-examine an algorithm.23 If an AI determines a specific multi- million dollar tax liability, but the proprietary nature of the software prevents the claimant from explaining exactly how the AI weighed the variables, opposing counsel is stripped of their rightto effectively test the evidence.24
B. THE SECTION 19 FLEXIBILITY VS. SECTION 65B MANDATE
This procedural dilemma is exacerbated by the conflicting statutory frameworks governing evidence in Indian-seated arbitrations. Section 19(1) of the Arbitration Act explicitly states that the arbitral tribunal “shall not be bound by the Code of Civil Procedure, 1908 or the Indian Evidence Act, 1872.” 25 This provision was designed to afford tribunals maximum procedural flexibility, theoretically allowing arbitrators to admit AI-generated audits without adhering to strict rules of authentication.
However, this flexibility is increasingly colliding with the stringent digital authentication mandates of Indian jurisprudence. An AI-generated tax audit is, fundamentally, an electronic record. Under Section 65B of the Evidence Act (and its successor, Section 63 of the Bharatiya Sakshya Adhiniyam, 2023), electronic records are entirely inadmissible unless accompanied by a mandatory certificate confirming that the computer system was operating properly and the data was not tampered with.26 The Supreme Court of India, in Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, reaffirmed the absolute necessity of this certificate for digital evidence.27
While Section 19 allows tribunals to bypass the strict application of the Evidence Act,
ignoring the principles of Section 65B when dealing with highly complex algorithmic evidence presents a massive risk. If a party submits an AI audit, who signs the Section 65B certificate? A legal counsel cannot truthfully certify the “proper operation” of a deep learning neural network they did not program. Conversely, the AI developer may refuse to disclose the underlying operational logic, citing trade secret protection.28
C. DUE PROCESS AND THE “FUNDAMENTAL POLICY OF INDIAN LAW”
If a tribunal chooses to rely heavily on Section 19 flexibility and admits a “black box” AI tax audit over the objections of opposing counsel, the resulting arbitral award becomes highly vulnerable to annulment. Section 34(2)(b)(ii) of the Arbitration Act allows an Indian court to set aside an award if it is in conflict with the “fundamental policy of Indian law.”29
The Supreme Court, in Ssangyong Engineering & Construction Co. Ltd. v. National
Highways Authority of India, definitively ruled that an award violates this fundamental policy if it is based on materials or evidence taken behind the back of the parties, without giving them an opportunity to comment or cross-examine.30 By definition, an inscrutable algorithmic output functions as “secret evidence” if the underlying methodology cannot be scrutinized by the opposing party. Furthermore, if an arbitrator relies entirely on the AI Auditor’s quantified number without comprehending the mathematical logic, the arbitrator arguably delegates their adjudicatory function to the machine, rendering the decision an “unreasoned award.” 31
Therefore, a rigid application of either extreme blanket admissibility under Section 19 or absolute inadmissibility under Section 65B is unworkable for modern commercial arbitration. To resolve this impasse, practitioners and institutions must construct a specialized, watertight evidentiary architecture tailored to algorithmic evidence.
IV. THE PRACTITIONER’S BLUEPRINT: A FIVE-POINT FRAMEWORK FOR ALGORITHMIC PROOF
The inherent clash between Indian statutory mandates (Section 65B of the Evidence Act) and procedural flexibility (Section 19 of the Arbitration Act) demonstrates that current frameworks are insufficient to handle the epistemological gap created by algorithmic evidence. To prevent the arbitrary exclusion of highly efficient AI tax audits and the consequent setting aside of awards under Section 34 practitioners must construct a specialized, watertight evidentiary architecture. The following five-point blueprint provides highly practical, prophylactic measures spanning the entire lifecycle of an international commercial arbitration.
A. PRE-DISPUTE: THE ‘ALGORITHMIC CONSENT AND TRANSPARENCY’ (ACT) CLAUSE
The most legally robust method to bypass the rigid certification mandates of Section 65B of the Indian Evidence Act is to neutralize the issue at the contractual stage via party autonomy. Corporate counsel drafting cross-border M&A agreements or complex supply contracts must transition away from standard boilerplate dispute resolution clauses and incorporate an ‘Algorithmic Consent and Transparency’ (“ACT”) clause.
An ACT clause explicitly stipulates ex ante that in the event of a dispute regarding the
quantum of a tax indemnity or transfer pricing adjustment, both parties consent to the use of Machine Learning or AI auditing tools. Crucially, to make this clause watertight against subsequent Section 34 “public policy” challenges, the clause must contain a mutual waiver: parties expressly agree that algorithmic outputs shall be admissible without a strict Section 65B certificate, provided that the deploying party complies with an agreed-upon transparency protocol (such as the SVAMC Guidelines).32 By contractually framing the evidentiary rules, parties harness Section 19 to legally insulate the AI evidence from domestic statutory hurdles.
B. DOCUMENT PRODUCTION: METHODOLOGY DISCLOSURE VIA “CONFIDENTIALITY
RINGS”
When a party pleads a complex tax claim supported by an AI Auditor, a severe conflict arises during the document production phase: the deploying party wishes to protect the AI’s proprietary source code as a trade secret, while the opposing party demands to see the algorithm’s logic to exercise their right to cross-examination.
To resolve this pragmatically, tribunals must issue a procedural order establishing a
“Confidentiality Ring” (a tool frequently used in Indian intellectual property litigation). The pleading party must submit an Algorithmic Methodology Report detailing the AI’s training datasets, error rates, and variable weights. If the opposing party challenges the algorithm’s underlying code, the code is disclosed only within the Confidentiality Ring accessible exclusively to the opposing party’s independent technical experts and legal counsel, who are bound by severe non-disclosure undertakings.33 This entirely preserves the AI developer’s trade secrets while guaranteeing the opposing party’s fundamental right to due process.
C. EVIDENTIARY PRESENTATION: THE “TANDEM EXPERT” SPONSORSHIP MODEL
Under Section 45 of the Indian Evidence Act, expert opinion is strictly limited to human persons possessing special skills; an algorithm cannot possess independent legal personality or act as a witness.34 Therefore, merely submitting a “Quantum Report” generated by an AI Auditor renders the evidence fundamentally defective and legally orphaned. To ensure practical admissibility, practitioners must utilize the “Tandem Expert” sponsorship model.35 The algorithmic evidence must be co-authored and sponsored by two human experts testifying in tandem: a Domain Expert (a traditional forensic accountant who testifies to the commercial tax principles and interprets the final quantum) and a Technical Expert (an AI Data Scientist who testifies to the proper functioning, lack of bias, and logical architecture of the algorithm). This bifurcated sponsorship guarantees that opposing counsel
has appropriate human targets to cross-examine regarding both the law (the tax calculation) and the mechanics (the algorithmic process), effectively curing the “black box” defect.36
D. TRIBUNAL ADJUDICATION: THE TRIBUNAL-APPOINTED TECHNICAL EXPERT
(TATE)
A practical reality of modern arbitration is that most arbitrators are eminent jurists, not software engineers. If both parties submit competing tax audits generated by different, highly
complex AI models, the tribunal is entirely unequipped to evaluate which algorithm is legally and mathematically sound, risking an “unreasoned award.”
To make the adjudication watertight, tribunals must actively invoke Section 26 of the
Arbitration Act, which empowers the tribunal to appoint its own experts.37 At the onset of a tech-heavy tax dispute, the tribunal should appoint a Tribunal-Appointed Technical Expert (“TATE”).38 The TATE’s sole mandate is not to decide the tax law, but to act as a technical translator for the arbitrators auditing the Algorithmic Methodology Reports submitted by the parties, testing for algorithmic bias, and advising the tribunal on whether the AI models meet the Daubert standard of scientific reliability.39
E. SYSTEMIC REFORM: INSTITUTIONAL ‘SAFE HARBOR’ PRESUMPTIONS
Finally, for long-term practical applicability, premier Indian arbitral institutions such as the Mumbai Centre for International Arbitration (“MCIA”) and the Delhi International Arbitration Centre (“DIAC”) must modernize their institutional rules. Relying on ad-hoc procedural orders is insufficient for global commercial certainty.
These institutions should draft an “Algorithmic Evidence Annexure” that creates an
evidentiary ‘Safe Harbor’. 40 The rule should state that if an AI Auditor complies with the transparency, human-oversight, and logging mandates outlined in the EU AI Act for “High-Risk Systems,” the algorithmic output enjoys a rebuttable presumption of admissibility and reliability in the arbitration.41 This shifts the burden of proof: instead of the deploying party struggling to authenticate the digital record under archaic Evidence Act standards, the opposing party must technically prove why the globally compliant AI is flawed.
V. CONCLUSION
The integration of the “AI Auditor” into international commercial arbitration represents
a permanent paradigm shift in how complex, multi-jurisdictional taxation claims are quantified, pleaded, and proven. As global corporate holding structures and cross-border M&A transactions become increasingly labyrinthine, the traditional human forensic accountant is no longer equipped to parse decades of transfer pricing ledgers with the necessary speed and accuracy. The reliance on machine learning algorithms to trace hidden tax liabilities is no longer merely a theoretical possibility; it is a present commercial necessity. However, as this paper has demonstrated, the deployment of such advanced technology introduces a profound evidentiary minefield for arbitrations seated in India. While the Supreme Court’s ruling in Vidya Drolia safely brings contractual tax indemnities within the jurisdictional competence of arbitral tribunals, the procedural mechanics of proving these claims remain severely fractured. The rigid application of the Indian Evidence Act’s digital authentication mandates, specifically the strict certification requirements of Section 65B, is functionally incompatible with the proprietary, “black box” nature of deep learning algorithms. Conversely, allowing arbitral tribunals unfettered discretion under Section 19 of the Arbitration Act to admit unexplainable algorithmic outputs eviscerates the opposing party’s statutory right to cross-examination. Ultimately, admitting an AI’s unverified calculation as concrete proof renders the final award highly susceptible to annulment under Section 34 for violating the fundamental policy of Indian law. To safely navigate this epistemological clash, the Indian arbitral regime must evolve
beyond the binary choice of rejecting technology or sacrificing due process. By looking beyond rigid statutory interpretations and enforcing structured transparency, Indian tribunals can bridge the gap between technological efficiency and procedural fairness.
The comprehensive Practitioner’s Blueprint proposed in this paper encompassing
‘Algorithmic Consent and Transparency’ (“ACT”) clauses, the utilization of Confidentiality Rings during document discovery, the bifurcated “Tandem Expert” sponsorship model, and the appointment of Tribunal-Appointed Technical Experts (“TATE”) provides a highly pragmatic, watertight framework for corporate counsel and arbitrators. Ultimately, adapting these evidentiary frameworks is imperative. If India is to solidify its standing as a premier, pro-arbitration jurisdiction on the global stage, its institutions and practitioners must proactively construct procedural “safe harbours” for algorithmic evidence. By enforcing structured transparency and human accountability, tribunals can ensure that the AI Auditor serves as a formidable instrument of truth, rather than a shield for opaque adjudication.
- Arbitration and AI, WHITE & CASE (June 2, 2025), https://www.whitecase.com/insight-our-thinking/2025-
international-arbitration-survey-arbitration-and-ai; Martin Magal, Alexander Calthrop & Katrina Limond, Artificial Intelligence in Arbitration: Evidentiary Issues and Prospects, A&O SHEARMAN (Jan. 12, 2024), https://www.aoshearman.com/en/insights/artificial-intelligence-in-arbitration-evidentiary-issues-and- prospects. ↩︎ - maxi Scherer, Artificial Intelligence in Arbitral Decision-Making: The New Enlightenment?, in ARBITRATION’S AGE OF ENLIGHTENMENT? 21 ICCA CONG. SER. 11, 14 (2023), https://cdn.arbitration-icca.org/s3fs-public/document/media_document/Congress_Series_no21_Arbitrations_Age_of_Enlightenment.pdf (last visited Feb. 19, 2026). ↩︎
- Gary B. Born, International Commercial Arbitration 2374 (3d ed. 2021). ↩︎
- The Arbitration and Conciliation Act, 1996, § 19(1), No. 26, Acts of Parliament, 1996 (India). ↩︎
- The Indian Evidence Act, 1872, § 65B, No. 1, Acts of Parliament, 1872 (India). See also Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 S.C.C. 1 (India). ↩︎
- The Arbitration and Conciliation Act, 1996, § 34(2)(b)(ii) (India); see also Ssangyong Eng’g & Constr. Co. v. Nat’l Highways Auth. of India, (2019) 15 S.C.C. 131, ¶ 54 (India). ↩︎
- See generally Chaitanya Chaturvedi, Taxation Turbulence: The Controversy of Arbitration in International Tax Disputes, IJAL BLOG (Mar. 3, 2025), https://www.ijal.in/post/taxation-turbulence-the-controversy-of-arbitration-
in-international-tax-disputes. ↩︎ - The Income Tax Act, 1961, No. 43, Acts of Parliament, 1961 (India), § 293. ↩︎
- Vidya Drolia v. Durga Trading Corp., (2021) 2 SCC 1. ↩︎
- Cairn Energy PLC & Cairn UK Holdings Ltd. v. Republic of India, PCA Case No. 2016-7, Final Award, ¶ 846
(Dec. 21, 2020). ↩︎ - Vidya Drolia, (2021) at 9. ↩︎
- Booz Allen & Hamilton Inc. v. SBI Home Finance Ltd., (2011) 5 SCC 532. ↩︎
- Vidya Drolia, (2021) at 9, ¶ 76. ↩︎
- Id; See also Tarun Jain, Tax Arbitration in India: Scope and Trends, SCC ONLINE BLOG (Mar. 5,2024), https://www.scconline.com/blog/post/2024/03/05/tax-arbitration-in-india-scope-and-trends/. ↩︎
- William W. Park, Arbitrability and Tax, in ARBITRABILITY: INTERNATIONAL AND COMPARATIVE PERSPECTIVES 179 (Loukas A. Mistelis & Stavros L. Brekoulakis eds., 2009), https://scholarship.law.bu.edu/faculty_scholarship/2287/ (last visited Feb. 20, 2026). ↩︎
- The Companies Act, 2013, No. 18, Acts of Parliament, 2013 (India), § 90; see also The Companies (Significant Beneficial Owners) Rules, 2018 (India). ↩︎