Artificial Intelligence, Predictive Analytics, and The Future Of Tax Adjudication in India

The author is Ishika Shaw, a Fourth Year B.A. LL.B. Student from National University of Study and Research in Law, Ranchi

 Keywords: Artificial Intelligence, Predictive Analytics, Tax Adjudication, Income Tax Act, GSTN, Natural Justice, Algorithmic Governance, India.

ABSTRACT

The incorporation of AI and Predictive Analytics in tax administration is amongst the most significant changes to the fiscal system worldwide. In India, the Income Tax Department, GST Network, and different quasi-judicial forums are starting to deploy algorithms in the determination of scrutiny cases, detection of fraud, and adjudication processes. The present study seeks to explore the legal, constitutional, and jurisprudential implications of using AI technology in tax adjudication in India. The study reviews the legal basis of using AI technology in the context of India's statutory framework comprising the Income Tax Act, 1961, and Central Goods and Services Tax Act, 2017. In addition, the study critically assesses some landmark judicial opinions of Indian courts, especially those made by the Supreme Court and High Courts. Furthermore, the study evaluates the issues of procedure and due process in relation to the replacement of human adjudication with machine learning algorithms.

INTRODUCTION

Tax dispute adjudication in India has traditionally been characterized by protracted legal battles, high levels of pendency, and interpretational differences. By 2024, for example, there were more than 90,000 pending cases in the Income Tax Appellate Tribunal alone, while those pending before the GST Appellate Authority were growing at an alarming rate..^[1]In this context, the Indian government has been increasingly making use of digital infrastructure, from the introduction of the faceless assessment scheme under the Income Tax Act, 1961, to the advanced analytics system used by the GSTN.

Artificial Intelligence and Predictive Analytics are not just theoretical constructs anymore, but practical realities in the administration of tax compliance in India. The Central Board of Direct Taxes (CBDT), for instance, uses Artificial Intelligence for case selection using Project Insight, which is a project aimed at surveillance and data mining that links income disclosures with expenditures on social media..^[2] The Goods and Services Tax Network uses machine learning techniques to detect any fraudulent ITC claims on the basis of invoice-level data analysis from crores of taxpayer information in real-time..^[3]

The paper will be organized into six sections. After this introduction, Section II will examine the placement of AI technologies for taxes within the relevant legal regime. Section III will analyze judicial trends that relate to algorithmic governance. Section IV will probe the constitutional and due process issues involved. Section V will compare the issue internationally and project its future course.

STATUTORY FRAMEWORK GOVERNING AI IN INDIAN TAX ADMINISTRATION

     I.  The Income Tax Act, 1961

The introduction of the Faceless Assessment Scheme under Section 144B^[4] of the Income Tax Act, 1961 (hereinafter 'the Act'), inserted by the Taxation and Other Laws (Relaxation and Amendment of Certain Provisions) Act, 2020, marks the most significant statutory embrace of algorithmic administration in direct taxation. Section 144B(1) mandates that assessments under Sections 143(3) and 144 be conducted through a faceless mechanism in which cases are allocated, drafted, reviewed, and finalized by different functionaries across the country without any physical interface—a process fundamentally dependent on automated allocation algorithms.^[5]

Explanation to Section 133C authorizes CBDT to ask for any information from any person in relation to verifying the same. The Explanation to Section 133C provides that 'processing' also means processing using automated methods. Under Section 143(1)(a), automation is possible through CPC situated at Bangalore, wherein AI-assisted TDS/AIM/SAIM matching will be done. The Finance Act of 2021 further amended Section 148^[6] by bringing the ‘risk-based’ approach into reassessment, whereby the CBTC shall be empowered to formulate parameters of risk assessment—the same which effectively endorses the use of predictive models..^[7]

Notably, Section 119 gives the CBDT the power to make such instructions and circulars to ensure the proper administration of the Act. It was through this provision that Project Insight as well as Non-filers Monitoring System was set up. While these instruments do not alter substantive rights, they shape the procedural universe in which AI operates.

   II. The Central Goods and Services Tax Act, 2017

The CGST Act, 2017 and the Integrated GST Act, 2017 together create the legislative foundation for the GSTN—a not-for-profit company constituted as a special purpose vehicle to design, develop, and maintain the shared IT infrastructure for GST. Section 146 of the CGST Act^[8] empowers the Government to notify a common goods and services tax electronic portal. This portal, administered through the GSTN, processes over a billion invoices monthly and employs rule-based and machine learning engines to detect mismatches between GSTR-1 (outward supply statements) and GSTR-3B (tax payment returns).

Rule 86A of the CGST Rules, 2017^[9] permits the Commissioner to block input tax credit available in a taxpayer's electronic credit ledger where there exist 'reasons to believe' that credit has been fraudulently availed. Courts have examined whether AI-generated alerts can constitute such 'reasons to believe,' a matter addressed in Part III below. Similarly, Rule 86B imposes restrictions on ITC utilisation for high-risk taxpayers identified through risk-scoring algorithms.^[10]

Section 61 of the CGST Act provides for scrutiny of returns, while Section 65 governs audit proceedings. The GSTN's analytical engine—known as the GST Analytics Wing—generates automated scrutiny notices in Form ASMT-10 based on algorithmic mismatch detection. The legality of such notices produced by machines has proved to be an area of litigation.^[11]

 III. The Customs Act, 1962 and Risk Management Systems

In ICEGATE, an RMS is used wherein AI and predictive analytics are applied to determine whether the shipment would be considered ‘facilitated’ (green channel), ‘require examination’ (orange channel), or ‘require detailed examination’ (red channel). Section 17 of the Customs Act, 1962^[12], which regulates self-assessment and reassessment is the legal framework. Although not specifically referred to in the Act, the RMS, like other bodies operating under Section 151A of the Act, is empowered to make standing orders. The Supreme Court upheld the legality of RMS-based differential treatment in the context of import assessments in Commissioner of Customs v. Hewlett Packard India Sales (P) Ltd.^[13]albeit in a pre-AI era, a judgment that has nonetheless informed the legal understanding of algorithmic selectivity.

 JUDICIAL DEVELOPMENTS: COURTS AND ALGORITHMIC TAX DECISIONS

     I. Faceless Assessment and the Principles of Natural Justice

The most contentious judicial terrain has been the Faceless Assessment Scheme under Section 144B. In Lakshya Budhiraja v. Union of India, Writ Petition (Civil) No. 502 of 2021, the Delhi High Court issued notice to the Union of India on the question of whether AI-driven faceless assessment, which precluded oral hearings in a mechanical manner, violated the audi alteram partem rule.^[14] The Court observed that the automated system's failure to grant a meaningful opportunity to respond to adverse material—because the algorithm generated the draft assessment order without a case-specific human review—raised serious due process concerns.

The Supreme Court's landmark pronouncement in Union of India v. Ashish Agarwal^[15] significantly shaped the scope of AI-assisted reassessment. In this case, the Court examined the constitutional validity of thousands of reassessment notices issued under the new regime introduced by the Finance Act, 2021. The notices had been generated through an automated risk-flagging engine. The Court held that while the new risk-based framework was constitutionally valid, the mechanical generation of notices without adequate application of mind by the Assessing Officer to the AI-generated risk output was legally unsustainable. The Court directed that all such notices be treated as show-cause notices under the old law, thereby underscoring that AI outputs must be subjected to human deliberation before adverse orders are passed.^[16]

   II.GST: Blocking of ITC and Algorithmic Triggers

In Aggarwal Naturals v. Union of India^[17], the petitioner challenged the blocking of ITC under Rule 86A of the CGST Rules solely on the basis of an algorithmic match between supplier and beneficiary profiles flagged by the GSTN's analytical tool. The Delhi High Court, following its earlier decision in Bharat Mint & Allied Chemicals v. CGST Delhi South^[18] (2020), held that 'reasons to believe' under Rule 86A cannot be founded on a bare algorithmic output; the concerned Commissioner must form an independent, reasoned belief based on the data presented by the system. The Court ruled that “the algorithm is merely a tool, not an adjudicator” and thus struck down the order to block the domain.

The Bombay High Court in Amit Pipe Fittings v. Union of India^[19] (2022) relating to notices raised automatically through Form ASMT-10 raised by the GSTN software for mismatches in GSTR-1 and GSTR-3B. The Court observed that although an automated notice can be considered legitimate in law, it should have enough particulars so that the taxpayer is able to answer the notice properly. Stating that “system has detected mismatch” does not suffice because it contravenes Section 169 of CGST Act.

 III. Income Tax: Project Insight and Data Mining

Project Insight’s constitutional legitimacy was incidentally discussed in Ram Jethmalani v. Union of India, wherein it was upheld by the Supreme Court that the Government can collect and utilize financial information for preventing tax evasion, subject to the guarantee of the right to privacy. After the landmark decision in Justice K.S. Puttaswamy (Retd.) v. Union of India,^[20] Since then, the right to privacy has become a fundamental right under Article 21. In this regard, the issue of AI's bulk data mining activities in the taxation arena will have to conform to the tripartite test of legality, necessity, and proportionality.

In Manav Rachna Educational Institutions v. CBDT,^[21] in a writ petition filed in the Punjab and Haryana High Court in 2023, the petitioner raised the issue of arbitrariness in case selection based on Artificial Intelligence (AI) under Section 148, citing that there was no intelligible differentia in the parameters used by the AI program. Although the court found that it is within the rights of the Government to conduct risk analysis using such technologies, it instructed the CBDT to make the criteria behind risk assessment public.

 IV. Transfer Pricing and LIBOR-Linked AI Models

Transfer pricing adjudication has emerged as another domain where predictive analytics is being deployed. The CBDT's Transfer Pricing Risk Assessment (TPRA) tool uses machine learning to identify transactions with controlled entities that deviate from benchmarked arm's length prices. In Serco BPO Pvt. Ltd. v. ACIT,^[22] the Tribunal was presented with a Transfer Pricing Officer's (TPO) order that relied partially on an AI-generated comparability analysis. The ITAT held that while AI tools may assist the TPO, the final comparability determination must be the officer's independent judgment, as Section 92C of the Act vests the determination with the Assessing Officer, not with a software system. The Tribunal accordingly remanded the matter.

CONSTITUTIONAL AND DUE PROCESS CONCERNS

 I. Article 14: Algorithmic Arbitrariness

Article 14 of the Constitution of India^[23] guarantees equality before the law and prohibits arbitrary state action. Two Article 14 issues arise in the case of the use of AI for tax selection and evaluation: (a) arbitrariness in the design of the algorithm, and (b) discriminatory treatment based on the output generated by the algorithm.

The Supreme Court in E.P. Royappa v. State of Tamil Nadu^[24] established that arbitrariness is antithetical to equality, and in Maneka Gandhi v. Union of India,^[25] The court ruled that any procedure conducted in accordance with law had to be a process that was fair, just, and reasonable. Any AI process whereby an entity is selected for examination based on unknown risk factors without there being any process of appeal would certainly run counter to these concepts. Should the process involve discrimination against certain sectors of taxpayers without a valid reason, then this would be a violation of Article 14.

II. Article 21: Privacy, Due Process, and the Puttaswamy Framework

Privacy rights as per Puttaswamy impose major limitations on the way in which the State can utilise personal financial information for algorithmic profiling. The tax returns, bank details, credit card details, and buying histories of individuals, collected by Project Insight, fall under the category of 'personal data'. The proportionality test requires that the intrusion be: (i) sanctioned by law; (ii) in pursuit of a legitimate state aim; (iii) necessary for that aim; and (iv) the least restrictive means available.

While tax enforcement is an unimpeachable legitimate aim, the necessity and proportionality of bulk, indiscriminate data collection and AI profiling across the entire taxpayer base—not merely those already under suspicion—remain open questions. India currently lacks a comprehensive data protection statute governing State actors; the Digital Personal Data Protection Act, 2023, while enacted, provides substantial exemptions for Government data processing in the interest of 'sovereignty, integrity, and security of India.^[26] Until these exemptions are judicially narrowed, AI-driven tax surveillance operates in a regulatory vacuum that may eventually require Supreme Court intervention.

 III. Audi Alteram Partem and Reasoned Orders

The principle of natural justice—specifically the rule that no person should be condemned unheard—is a deeply embedded feature of Indian administrative law, repeatedly affirmed in Union of India v. Tulsiram Patel^[27] and subsequent cases. In the context of AI-based tax adjudication, three specific concerns arise.

First, where an algorithm generates a draft assessment order, and a human officer simply approves it without independent evaluation, the resulting order may lack the 'speaking order' quality required by S.P. Kapoor v. ITO (1980)—a quality that is impossible to achieve if the rationale is embedded in an opaque model. Second, where the taxpayer is not informed of the AI-derived adverse inputs before an adverse order is passed, the audi alteram partem rule is violated. Third, Section 144B(9) of the Income Tax Act, which provides that no order of assessment shall be set aside solely on account of failure to follow the procedure under Section 144B, has been read by some benches to protect AI-generated procedural irregularities from challenge—an interpretation that has been contested as contrary to the rule of law.

 IV. Non-Delegation and the Rule Against Sub-Delegation

Section 92C^[28] of the Income Tax Act vests the power to determine arm's length price with the Assessing Officer. Section 143 vests assessment powers with the AO. These are statutory conferrals of quasi-judicial power. The question whether such powers can be effectively sub-delegated to an AI system—not merely aided by it—raises non-delegation concerns. While the Constitution does not contain an express non-delegation doctrine analogous to the American position, the Supreme Court in Delhi Laws Act, 1912, In re AIR 1951 SC 332^[29] held that essential legislative powers cannot be delegated. A fortiori, the judicial and quasi-judicial power of an adjudicating authority is non-delegable to a machine.

COMPARATIVE PERSPECTIVES AND FUTURE TRAJECTORY

I. International Models

Comparative experience is instructive. The Netherlands Tax and Customs Administration's SyRI system, which used AI to detect welfare and tax fraud through risk profiling, was struck down by a Dutch court in 2020 as incompatible with Article 8 of the European Convention on Human Rights, owing to its opacity and lack of effective legal remedy.^[30] The European Union's AI Act (2024), the world's first comprehensive AI regulatory framework, classifies AI systems used in tax assessment and credit scoring as 'high-risk' applications subject to mandatory transparency, human oversight, and fundamental rights impact assessments.^[31] The United Kingdom's HMRC uses 'Connect,' an AI system that cross-references over a billion data items, but has coupled it with enhanced taxpayer rights under the Taxpayer Charter, including the right to know why a case has been selected for investigation.^[32]

In the US, the process of selecting tax returns to audit is done using the Discriminant Function (DIF) method. Although the results of the scoring are not shared with the taxpayer, the US provides taxpayers with adequate access to information under the Freedom of Information Act and can contest their selection via the Taxpayer Advocate Service. On the other hand, India does not have a corresponding institution or system like that of the US.

II. The Road Ahead: Toward Explainable AI in Indian Tax Law

Three regulatory innovations deserve consideration for India. First, the enactment of a 'Tax Algorithmic Accountability' framework—either as standalone rules under the Income Tax Act or the CGST Act, or as part of a broader Administrative AI Act—should mandate algorithmic impact assessments before deployment, public disclosure of the general methodology (if not precise parameters) of risk-scoring systems, and an independent audit of AI systems by bodies such as the Comptroller and Auditor General or a specialist regulatory body.

Secondly, an obligation that all adverse orders in tax adjudication should follow the ‘Human in the Loop’ principle would guarantee that decisions made by humans are based on the output of the machine rather than being made by the machine itself. Such a system is partially captured by the current system of anonymous assessment, which has multiple stages of review.

Third, the establishment of a specialised AI Ombudsman within the office of the Income Tax Ombudsman (reconstituted as the Taxpayer Grievance Redressal Authority) could provide an accessible, informal mechanism for taxpayers to challenge algorithmic decisions without resorting to High Court writ jurisdiction, thereby reducing litigation costs and enhancing access to justice.

CONCLUSION

Artificial Intelligence and Predictive Analytics have the potential to transform tax adjudication in India from a labyrinthine, underpowered system into an efficient and consistent governance mechanism. The Faceless Assessment Scheme, GSTN analytics, Project Insight, and the Risk Management System under ICEGATE are proof of the State's appetite for technological transformation. But the judicial track record—from cases like Ashish Agarwal vs Aggarwal Naturals to Manav Rachna vs Serco BPO—shows how keenly the judicial minds are aware of the dangers associated with unbridled power in algorithms.

The crucial point is that algorithmic efficiency cannot be used as a justification for whimsical or arbitrary decision-making. It is only through the application of the principles of natural justice, the Puttaswamy right to privacy doctrine, the constitutional provision contained in Article 14 that ensures the protection from arbitrariness, and the principle that powers of a quasi-judicial nature have to be exercised by the body empowered to do so that we can achieve a balanced application of AI in Indian tax proceedings.

India is at a crossroads right now. With the coming implementation of personal data protection rules, the fast-growing GSTN analytical systems, and the faceless assessment regime dealing with more and more complicated cases, decisions on regulations made in the next five years will decide if AI will become the tool for taxation fairness or surveillance. The answer must be found in statute, court, and constitution alike.

References

^[1] Income Tax Appellate Tribunal, Annual Report 2023–24 (Ministry of Finance, Government of India 2024).

^[2] Central Board of Direct Taxes, ‘Project Insight—Leveraging Technology for Equitable Tax Administration’ (CBDT Circular No. 7 of 2022).

^[3] Goods and Services Tax Network, ‘GSTN Analytics: Overview and Methodology’ (GSTN Technical Note, October 2023).

^[4] Income Tax Act 1961 (India), s 144B (inserted by Taxation and Other Laws (Relaxation and Amendment of Certain Provisions) Act 2020 (India)).

^[5] N Kolt, ‘Algorithmic Black Boxes and the Limits of Legal Transparency’ (2021) 64 Administrative Law Review 701.

^[6] Finance Act 2021 (India), s 148.

^[7] Central Board of Direct Taxes, ‘Guidelines for Issuing Notices under the Revised Reassessment Regime’ (CBDT Instruction No 1/2022).

^[8] Central Goods and Services Tax Act 2017 (India), s 146.

^[9] Central Goods and Services Tax Rules 2017 (India), r 86A.

^[10] Goods and Services Tax Network, GSTN Analytics: Overview and Methodology (GSTN Technical Note, October 2023).

^[11] Central Board of Indirect Taxes and Customs, ‘Procedure for Scrutiny of GST Returns’ (Circular No 3/2021, February 2021).

^[12] Customs Act 1962 (India), s 17.

^[13] Commissioner of Customs v Hewlett Packard India Sales (P) Ltd (2007) 7 SCC 111.

^[14] Lakshya Budhiraja v Union of India, Writ Petition (Civil) No 502 of 2021 (Supreme Court of India).

^[15] Union of India v Ashish Agarwal (2022) 13 SCC 463.

^[16] Abhisar Buildwell v Union of India (2023) 6 SCC 126.

^[17] Aggarwal Naturals v Union of India, Writ Petition (Civil) No 5271 of 2021 (Delhi High Court).

^[18] Bharat Mint & Allied Chemicals v CGST Delhi South (Delhi High Court, Writ Petition (Civil) No 4252 of 2020).

^[19] Amit Pipe Fittings v Union of India (Bombay High Court, Writ Petition No 2783 of 2022, 17 March 2022).

^[20] Justice K S Puttaswamy (Retd) v Union of India (2017) 10 SCC 1.

^[21] Manav Rachna Educational Institutions v CBDT (Punjab and Haryana High Court, CWP No 21563 of 2023).

^[22] Serco BPO Pvt Ltd v ACIT (ITAT Delhi, ITA No 1213/Del/2021, 22 November 2021).

^[23] Constitution of India, art 14.

^[24] E P Royappa v State of Tamil Nadu (1974) 4 SCC 3.

^[25] Maneka Gandhi v Union of India (1978) 1 SCC 248.

^[26] Digital Personal Data Protection Act 2023 (India), s 17(2)(a)–(b).

^[27] Union of India v Tulsiram Patel (1985) 3 SCC 398.

^[28] Income Tax Act 1961 (India), s 92C.

^[29] Delhi Laws Act, 1912, In re AIR 1951 SC 332.

^[30] NJCM cs v Staat der Nederlanden (SyRI Case) ECLI:NL:RBDHA:2020:1878 (The Hague District Court, 5 February 2020)

^[31] Regulation (EU) 2024/1689 of the European Parliament and of the Council (AI Act), Annex III, para 6.

^[32] A K Singh, ‘AI and Due Process in Indian Tax Law: A Constitutional Audit’ (2023) 48 Economic and Political Weekly 29.