AI in GST Administration: Why Algorithmic Tax Enforcement Is a Threat Without Foundational Reforms

Introduction: A Digital Promise Still Unfulfilled

When the Goods and Services Tax was rolled out across India, the central promise was transformative — a unified, technology-driven tax framework that would curtail human discretion, foster transparency, and ease the compliance burden for honest assessees. Years down the line, the reality for a significant portion of small and medium enterprises has been far less reassuring. Persistent portal failures, relentless reconciliation demands between GSTR‑1, GSTR‑3B, GSTR‑2A/2B and GSTR‑9, and a steady stream of mechanically generated notices that bear little connection to actual business operations have become routine grievances.

Against this backdrop, the tax administration has now chosen to introduce yet another layer of complexity: Artificial Intelligence and sophisticated data analytics. Official communications and policy advocates frequently describe AI as a near-magical remedy for compliance failures, fraud detection, and faster assessments. However, deploying aggressive AI-driven enforcement over an already unstable technological and institutional foundation risks transforming a tool meant to support fair administration into an instrument that actively endangers assessee rights. This article critically examines the role AI is being assigned within GST administration, the compounded risks arising from existing data and portal infirmities, the growing judicial unease with machine-generated orders, and the essential safeguards that must precede any meaningful AI integration.

How AI Is Being Deployed in GST Enforcement

The integration of AI into GST administration broadly operates across three functional domains:

1. Risk Profiling and Algorithmic Scoring

Tax officers are now furnished with algorithmically generated risk scores and red-flag registers derived from cross-referencing GST return data, e-way bill records, e-invoice databases, banking information, and other external repositories. These scores determine which assessees attract scrutiny and which do not — often without any human officer exercising independent judgment at the threshold stage.

2. Pattern Detection and Fraud Analytics

Large-scale analytical engines are deployed to mine supply chains for indicators of fake invoicing, circular trading arrangements, and shell entity networks. Invoice-wise data extracted from GSTR‑1 is routinely compared against summary figures declared in GSTR‑3B. Similarly, Input Tax Credit (ITC) flows are tracked from supplier to recipient across multiple tiers to detect possible evasion patterns.

3. Automated Discrepancy Reporting

Systems auto-generate discrepancy alerts when figures in GSTR‑1 diverge from GSTR‑3B, or when the ITC claimed in GSTR‑3B does not reconcile with the credit available in GSTR‑2A/2B, or when annual returns filed in GSTR‑9/9C are inconsistent with monthly return data. These reports feed directly into the notice-generation machinery.

Critical Observation: The effectiveness of any AI system is intrinsically bounded by the quality of data it consumes and the quality of human interpretation applied to its outputs. Where both dimensions are compromised — as they currently are in the GST ecosystem — the deployment of AI amplifies errors rather than resolving them.

The Fragile Foundation: GST Portal Deficiencies That Predate AI

Before examining AI-specific risks, it is essential to appreciate the structural deficiencies that have plagued the GST framework since its inception in 2017. These pre-existing vulnerabilities form the contaminated data environment into which AI systems are now being introduced.

Persistent Reconciliation Failures

  • Chronic mismatches between GSTR‑1 and GSTR‑3B have generated waves of notices and protracted litigation, even where no actual tax evasion existed.
  • Divergences between GSTR‑2A/2B and GSTR‑3B arise routinely because suppliers either fail to file or file incorrect returns — consequences that fall squarely on the recipient assessee, who has already remitted full tax to the supplier.

Unreliable Portal Infrastructure

  • The GSTN portal has exhibited repeated technical failures including inconsistent GSTR‑2A data, invoice viewing limitations, session timeouts, connectivity disruptions, and inability to generate reports for extended periods.
  • Downloaded Excel data frequently diverges from portal-generated reports, creating genuine uncertainty about which dataset carries legal authority.

Registration and Identity Vulnerabilities

  • Non-existent or fraudulently registered dealers regularly appear within supply chains, ultimately resulting in ITC denial for legitimate, unsuspecting buyers.

Accessibility Barriers

  • For assessees operating from semi-urban or rural locations, poor internet infrastructure makes the portal experience particularly hostile, with routine compliance tasks consuming disproportionate time and resources.

These are not peripheral technical inconveniences. Each of these deficiencies directly distorts the computation of tax liability, interest obligations, and penalty exposure. When such a flawed data environment is handed to an AI system as its raw input, the algorithm will faithfully reproduce and magnify every inconsistency it encounters. Machines do not possess contextual judgment — absent specific contrary programming, a portal-generated mismatch will be read by the system as potential evasion, regardless of its actual cause.

The Menace of Algorithmic Enforcement: Real-World Scenarios

The phenomenon increasingly described as "automated tax terrorism" — where opaque algorithms generate adverse outcomes for assessees without intelligible explanation — is a documented global concern, and the GST context renders Indian assessees particularly vulnerable.