India AI Tax Surveillance: Inside the ₹70,000 Crore Algorithm

In February and March 2026, the Indian state quietly executed one of the largest algorithmic tax crackdowns in global history. By analyzing 60 terabytes of billing data, authorities uncovered a massive ₹70,000 crore tax evasion scam within the restaurant industry. The operation did not rely on traditional tip-offs or manual audits; it was driven entirely by big data and Generative AI mapping Goods and Services Tax (GST) numbers across 1.77 lakh restaurant IDs.

The system detected ₹13,317 crore in post-billing invoice deletions alone. Analysts estimate that algorithmic sampling flagged between 25% to 27% of restaurant sales as suppressed nationwide.

Mainstream coverage has overwhelmingly focused on the sheer scale of the recovery and the technological prowess of the state. However, beneath the staggering numbers lies a profound shift in the relationship between the citizen and the state. India has built a highly efficient, data-driven tax surveillance apparatus. Yet, this algorithmic governance system operates as a black box—lacking transparency, statutory audit mechanisms, and a dedicated appeal process for AI-driven decisions.

The Architecture of Surveillance: Insight and ADVAIT

The foundation of this digital dragnet was laid in July 2016, when the Income Tax Department signed a contract with L&T Infotech to buildProject Insight. Estimated to cost ₹1,000 crore ($150–156 million), the massive data warehousing and AI platform was officially launched by the Central Board of Direct Taxes (CBDT) in 2017 to track direct tax evasion. By 2019, Project Insight was fully operational, integrating traditional financial data with social media scraping to build 360-degree taxpayer profiles.

Following the CBDT's lead, the Central Board of Indirect Taxes and Customs (CBIC) rolled outProject ADVAIT(Advanced Analytics in Indirect Taxation) in 2021. ADVAIT was designed to track GST, customs evasion, and supply chain anomalies.

The results were immediate and aggressive. Official data shows that between April and December 2023, AI tools like ADVAIT and BIFA helped register 14,597 GST evasion cases. In one instance in 2023, ADVAIT detected ₹11,000 crore in Integrated GST (IGST) evasion among just 24 large importers.

"Project Insight has fundamentally altered the risk-reward equation for potential tax evaders," stated a senior CBDT official in a September 2023 interview. "The system's ability to correlate data across multiple domains... means that discrepancies between reported income and actual economic activity become increasingly difficult to conceal."

Historically, this shift from manual tax collection to automated surveillance mirrors the implementation of the Permanent Account Number (PAN) system, but at a scale that fundamentally alters the regulatory landscape. It is heavily modeled after the United Kingdom's 'Connect' system, which cost £100 million and prevented £4.1 billion in tax losses. However, unlike the UK, Australia, or Canada—which typically pair algorithmic flagging with strict legislative safeguards—India's deployment has outpaced its legal frameworks.

The Black Box and the Inverted Burden of Proof

Official claims maintain that AI systems like ADVAIT and Project Insight provide "non-intrusive" tax administration that promotes voluntary compliance without harassment. The state argues that these systems are highly accurate, targeting only high-risk entities through advanced data matching and predictive analytics.

The available evidence, however, paints a different picture. AI systems frequently misread legitimate financial complexity as evasion. The reliance on automated "NUDGE" SMS and email campaigns often results in mass-flagging. Honest taxpayers are forced to spend significant resources proving their innocence against a black-box algorithm.

Corporate lawyers warn of the collateral damage. Legal experts at Shardul Amarchand Mangaldas & Co. note, "While the use of business intelligence has its share of benefits, care should also be taken not to be reliant only on the system, as sometimes false positives can also be generated." They advise that "the idea of the day should be the effective use of hands-on administration coupled with the electronic data analysis."

On the ground, legitimate businesses face severe hidden costs. They are forced to hire specialized consultants to reconcile their internal ERP data with the government's automated GSTN and Project Insight portals. When false flags are generated by the AI, the consequences are immediate and punitive: frozen input tax credits (ITC) and blocked bank accounts, which can paralyze the cash flow of small and medium enterprises.

Ironically, the "faceless assessment" system—originally designed to reduce human corruption—has removed the human element necessary to understand nuanced, legitimate business anomalies. The burden of proof has been inverted; taxpayers are effectively treated as guilty by the algorithm until they can digitally prove their innocence.

A Legal Vacuum: Privacy and Algorithmic Accountability

There is a glaring contradiction between India's aggressive push for "Ease of Doing Business" and the deployment of an opaque system where there is no dedicated statutory appeal mechanism for AI-driven decisions. The state has effectively privatized regulatory oversight by forcing businesses to bear the heavy compliance costs of aligning with the government's AI models.

In the landmarkK.S. Puttaswamy v. Union of Indiacase, the Supreme Court recognized informational privacy as a fundamental right under Article 21 of the Constitution. Yet, the legislative follow-through has provided massive loopholes for state surveillance.

The Digital Personal Data Protection (DPDP) Act, passed in August 2023, contains broad exemptions for government processing and entirely lacks provisions for algorithmic impact assessments. India currently has no statutory body to audit algorithmic systems, investigate AI harms, or issue binding directions on automated administrative decisions.

Privacy advocates and legal scholars argue that this brand of algorithmic governance creates "new risks to dignity and autonomy." Advocates warn that "the poorest and least digitally literate can be quietly locked out of entitlements by a faceless system, with little remedy." Because the legal framework relies on post-hoc liability rather than ex-ante authorization, citizens can only challenge algorithmic harms after their accounts have been frozen or their businesses disrupted.

The Future of Automated Statecraft

The recovery of ₹70,000 crore in suppressed turnover is an undeniable testament to the raw power of algorithmic enforcement. The Indian state has successfully built a digital panopticon that makes large-scale tax evasion mathematically unsustainable.

However, efficiency is not a substitute for justice. By deploying systems like Project Insight and ADVAIT without concurrent algorithmic transparency laws, the state has created a paradigm where the algorithm is the ultimate arbiter of financial truth. Until taxpayers are granted the right to know the parameters and risk-scoring models that judge them, India's AI tax regime will remain a highly lucrative, yet legally precarious, black box.