AI electoral roll cleansing risks wrongful voter exclusions in Telangana, experts warn

During the ongoing Special Intensive Revision (SIR) enumeration form distribution exercise in Telangana, researchers and policy experts have warned that relying solely on artificial intelligence (AI) to identify discrepancies in draft electoral rolls could lead to wrongful and large-scale voter exclusions.

Speaking on the sidelines of the discussion titled "SIR in Telangana: Data Cleansing vs. Social Welfare," Sabir Ahamed, lead at the Kolkata-based public policy and research organisation Sabar Institute, stated that human intervention is essential to correct errors flagged by AI systems. The discussion was organised by the Centre for Development Policy and Practice.

Ahamed warned that an over-reliance on technology and AI algorithms can result in a disaster, pointing to the opacity with which these systems have been functioning. To highlight the risks, he drew on patterns observed during the 2026 draft electoral rolls exercise in West Bengal.

According to Ahamed, the West Bengal data showed that unmapped voter errors largely affected marginalised Hindu communities. Meanwhile, logical discrepancies, such as spelling mismatches and electors having more than six siblings, became tools for exclusion that disproportionately affected Muslim communities.

The analysis in West Bengal revealed that constituencies with higher Muslim populations had lower rates of unmapped voters. However, these same constituencies recorded the highest number of cases currently under adjudication.

Ahamed explained that Telangana must draw lessons from these outcomes. He noted that when notices are served and responses must be filed, a one-size-fits-all approach is counterproductive. In West Bengal, many electors who received notices had to rely on local cyber cafes, where they were provided with generic template responses.

Furthermore, Ahamed stressed that social audits are a non-negotiable component of the process. Although the Election Commission mandates that every list must be published at the booth level, Ahamed stated this is usually not happening. He urged that the format of these lists must be accessible, searchable, and readable, rather than being placed behind excessive security features.