The rise of AI in GST enforcement: Are algorithms becoming the new tax officers?

India’s GST regime, implemented on July 1, 2017, is one of the most ambitious exercises in digital tax administration globally. The GSTN processes over one billion invoices monthly, and the data generated from e-invoicing, e-way bills, and returns has invited the deployment of artificial intelligence (AI) and machine learning (ML) for enforcement and compliance verification. What was once governed exclusively by the trained mind of a tax officer exercising quasi-judicial authority is now increasingly mediated by algorithms—raising a fundamental question: are algorithms becoming the new tax officers, and what safeguards must govern their operation?

The evolution of technology in Indian GST administration

The GST system was conceived as a “technology-first” regime, mandating electronic filing, digital invoice matching, and real-time goods movement reporting. Mandatory e-invoicing (introduced from October 2020 and extended to businesses exceeding Rs 5 crore turnover by August 2023) created a machine-readable dataset of unprecedented granularity. The initial years relied on rule-based mismatch identification, but as data volumes grew, CBIC and GSTN moved towards AI/ML models capable of detecting complex evasion patterns. By 2024-2025, enforcement actions relied substantially on algorithmically generated risk assessments.

Key deployments include the Business Intelligence and Fraud Analytics (BIFA) tool, which integrates data from GSTR-1, GSTR-3B, e-way bills, e-invoices, income tax returns, customs data, and bank information to generate risk scores and identify fraudulent networks through network analysis algorithms. The special drive against fake registrations (May 2023-2024) identified over 29,000 bogus GST registrations involving fraudulent ITC claims of approximately ₹25,000 crore, largely through AI models detecting suspicious registration patterns. Additionally, AI analyses e-way bill data for circular trading, distance anomalies, and temporal inconsistencies, while algorithmic risk scoring now drives the selection of returns for scrutiny under Section 61 of the CGST Act, replacing the earlier system of random or officer-initiated selection. GSTN has also introduced mandatory Aadhaar-based biometric verification and in-person verification at designated GST Suvidha Kendra for algorithmically identified high-risk registration applications.

Legal and constitutional concerns

Article 14 of the Constitution guarantees equality before the law. If an AI model trained on historically biased data perpetuates discriminatory enforcement patterns, it may violate the constitutional guarantee of non-arbitrary state action. The jurisprudence on Article 14 particularly the requirement of ‘intelligible differentia’ demands that algorithmic classifications be demonstrably rational.

Where algorithmic risk-flagging triggers adverse consequences such as suspension of registration under Rule 21A of the CGST Rules, blocking of ITC under Rule 86A, or proceedings under Sections 67 or 74 of the CGST Act, taxpayers are entitled to respond effectively. Yet they often cannot, because the algorithmic basis is opaque. A show-cause

notice stating ‘the system has identified you as high-risk’ without disclosing parameters does not satisfy natural justice requirements. Moreover, the statutory ‘reason to believe’ required under Sections 67 and 74 of the CGST Act must be that of the proper officer, not a machine as mechanical reliance on system-generated alerts without independent application of mind renders proceedings vulnerable to challenge.

For small traders, suspension of GST registration effectively halts business operations, and remedies through appeals or High Court are expensive and practically inaccessible for many affected taxpayers.

Global comparisons

India’s deployment of AI in tax enforcement is remarkable in scale but not unique. The EU’s AI Act, 2024 classifies AI in tax administration as ‘highly risky’, mandating transparency and human oversight. The OECD’s 2024 report recommends algorithmic transparency and impact assessments. The common thread across mature jurisdictions is the recognition that AI in tax enforcement requires specific governance frameworks, a recognition India must urgently address.

The way forward: Towards an AI governance framework for tax administration

The CGST Act should be amended to specifically address algorithmic decision-making, defining its permissible scope, mandating disclosure requirements, and establishing oversight mechanisms. CBIC should publish Algorithmic Impact Assessments examining model accuracy, false-positive rates, and potential for bias. The proper officer must demonstrably apply judicial mind and record reasons for agreeing or disagreeing with algorithmic recommendations, with a clear accountability framework for errors.

Conclusion

The deployment of AI in Indian GST enforcement is already a reality shaping the compliance experience of millions of taxpayers. The efficiency gains are undeniable with over 29,000 bogus registrations identified, ITC fraud networks disrupted, and billions of transactions analysed in real time.

However, efficiency cannot be the sole metric of a just tax system. Article 14’s guarantee of non-arbitrary state action, natural justice principles, and the statutory ‘reason to believe’ requirement are fundamental protections that must be adapted to algorithmic governance. India has an opportunity to lead the developing world in crafting a balanced framework, one that harnesses technology while remaining anchored in the rule of law. The question is not whether algorithms will play a role in tax enforcement—they already do. The question is how the government, through legislative intervention, will work towards making the role of algorithms subordinate to that of tax officers in tax governance and also align with constitutional values and taxpayer rights.

(The author is Senior Partner, S&A Law Offices. Views are personal.)