West Bengal Voter Deletion: How AI Erased 9.1 Million Voters

Ahead of the April 2026 Assembly elections, the Election Commission of India (ECI) removed approximately 9.1 million names from West Bengal’s electoral rolls, according to official ECI data. While the institution has framed this mass deletion as a necessary administrative cleanup, a closer examination of the Special Intensive Revision (SIR) reveals a heavy reliance on opaque algorithmic deduplication processes.

The deployment of automated verification software has raised profound questions about systemic flaws, the penalization of clerical data-entry errors, and the shifting of the legal burden of proof onto vulnerable citizens. By shrinking West Bengal's total electorate by nearly 12%—dropping the voter base from 7.66 crore to 6.75 crore—the ECI has executed one of the largest mass voter deletions in recent democratic history.

This is not merely a story of bureaucratic housekeeping; it is a systemic decode of how algorithmic governance, when deployed without adequate safeguards, can disproportionately disenfranchise millions.

The Architecture of the Algorithmic Dragnet

The mass deletion was not an overnight event but the result of a phased, technology-driven revision process that began outside of the ECI's regular electoral review cycles. Initiated in July 2025, the SIR process officially commenced in West Bengal by November 2025.

According to credible reporting, the ECI bypassed standard deduplication modules used in earlier state revisions. Instead, the commission shifted to legacy mapping software for West Bengal, attempting to match current voters to electoral rolls dating back to 2002-2004. This software relied heavily on artificial intelligence and automated algorithms to flag inconsistencies between the current voter list and legacy documents.

Millions of voters were subsequently placed into an "Under Adjudication" category based on what the software termed "logical discrepancies," as verified by official ECI statements.

The Anatomy of a "Logical Discrepancy"

What constitutes a logical discrepancy to an algorithm is often a routine reality for an Indian citizen. The algorithmic flags were triggered by minor spelling differences, title changes, and age mismatches.

During hearings in February 2026, the Supreme Court of India explicitly criticized the AI-driven software's parameters, noting that its standards were "not based on ground realities." For example, the AI flagged families with more than six children or cases involving underage marriage as inherent data anomalies, treating complex socio-economic realities as fraudulent registrations.

Furthermore, the system ruthlessly penalized voters for clerical data-entry errors made by state officials. Minor typographical variations across decades of paperwork were treated as evidence of illegitimacy.

"These are data-entry errors made by the officials. How is it my fault? I am being robbed of my constitutional right to vote over such silly reasons." —Faridul Islam, a 40-year-old voter deleted because legacy documents formatted his name as "Fa. Ridul" instead of "Faridul".

Data Breakdown: Error Rates and Demographic Disparities

The sheer scale of the deletions exposes the fragility of the ECI's automated deduplication system. When subjected to judicial scrutiny, the software's accuracy proved highly questionable.

Of the approximately 6.006 million voters flagged by the software specifically for "logical discrepancies," judicial review found that 3.26 million—over 54%—were actually eligible citizens. These individuals were ultimately retained on the rolls, while the remaining 2.71 million in this category were deleted. Analysts estimate that this 54% false-positive rate highlights a catastrophic inaccuracy in the automated flagging system, raising severe doubts about the integrity of the broader 9.1 million total deletions confirmed by the ECI in April 2026.

Disproportionate Geographic Impact

The algorithmic dragnet did not fall evenly across the state. Official data reveals staggering demographic and geographic disparities in the deletion rates:

• Nadia District: Recorded an unprecedented 77.86% deletion rate among flagged voters.

• North 24 Parganas: Saw 55.08% of flagged voters removed.

• Urban Divides: Kolkata North experienced a nearly 64% deletion rate, starkly contrasting with the 36.19% rate in Kolkata South.

• Absolute Volume: Murshidabad recorded the highest absolute number of deletions under the logical discrepancy category.

Historically, with roughly 11.6% of its pre-SIR electoral roll removed, West Bengal recorded the third-highest deletion rate among the nine Indian states subjected to this specific revision process, trailing only Gujarat and Chhattisgarh.

Inverting the Burden of Proof

Article 326 of the Indian Constitution guarantees universal adult franchise, establishing that voter deletion is a direct threat to representational legitimacy. Under Section 22(c) of the Representation of the People Act, 1950, and Rule 21, the state cannot strip a citizen of their voting rights without a fair, transparent hearing and an opportunity to respond.

However, the algorithmic approach of the SIR effectively inverted the legal burden of proof. Instead of the state bearing the responsibility to prove a voter was ineligible, millions of citizens summarily flagged by the software were forced to proactively prove their citizenship and identity.

This required navigating a bureaucratic maze, presenting cases before roughly 700 judicial officers and 19 special appellate tribunals. For vulnerable demographics lacking the time, financial resources, or digital literacy to appeal, an algorithmic flag became a final sentence of disenfranchisement.

The crisis reached a breaking point in April 2026 when the Supreme Court froze the Phase 1 electoral rolls ahead of the April 23 elections. While intended to halt further unchecked deletions, this freeze effectively locked out millions of already-deleted voters whose appeals remained pending in the backlogged tribunal system.

Civic Hygiene or Systemic Disenfranchisement?

The Election Commission maintains that the automated deduplication is a standard administrative necessity. Defending the SIR as a "routine act of civic hygiene," the ECI argues the process was designed to cleanse a bloated database of duplicates, deceased individuals, and shifted voters.

"The revision exercise has been carried out in a phased and transparent manner," a senior EC official stated, pointing to the public release of district-wise data as proof of complete accountability.

Yet, digital rights advocates and legal experts argue that the ECI's reliance on untested, opaque software violates the fundamental principles of natural justice. The rushed nature of the exercise prioritized algorithmic efficiency over constitutional due process. When a system operates with a 54% false-positive rate, it ceases to be an instrument of civic hygiene and becomes an engine of systemic exclusion.

The Future of Algorithmic Governance

The deletion of 9.1 million voters in West Bengal serves as a grim case study in the dangers of unchecked algorithmic governance. When institutions deploy automated systems to manage democratic rights, the opacity of the algorithm shields the institution from immediate accountability.

If a human official disenfranchised millions based on typographical errors like "Fa. Ridul," the systemic failure would be obvious and actionable. But when an algorithm does the same, it is shielded under the guise of "logical discrepancies" and technological objectivity.

As India moves deeper into an era of digitized governance, the ECI's deduplication process in West Bengal must be critically examined. The right to vote cannot be contingent on a citizen's ability to survive an algorithmic dragnet that fails to comprehend the ground realities of the very people it is meant to count.sq