Algorithmic Voter Management in India: Software, Transparency, SIR, and the ECI’s Refusal to Disclose Details
PoliticalMarketerIn shortAlgorithmic voter management in India refers to the use of digital databases, automated matching systems, analytical rules, artificial intelligence, and software-assisted verification to prepare and maintain electoral rolls.
On this page15 sections
The Election Commission of India (ECI) uses systems including ERONET and ECINET to process voter applications, compare demographic and photographic records, identify possible duplicate entries, support field verification, and assist Electoral Registration Officers (EROs). The issue gained national attention because software-generated flags increasingly affect who must undergo additional verification during the Special Intensive Revision (SIR), while the ECI declined in October 2026 to disclose the name, version, and detailed workings of the software used to identify certain electoral-roll anomalies.
How India Moved From Computerised Rolls to Algorithm-Assisted Voter Management
India's electoral-roll technology developed over almost three decades. The shift began as basic computerisation, moved to common national databases and online form processing, and later added automated matching and AI-supported identification of similar voter records.
Nationwide computerisation of electoral rolls began in 1997. Photographs were later incorporated into rolls nationwide ahead of the 2009 Lok Sabha election. These Photo Electoral Rolls created the data foundation needed for later photographic comparison systems.
ERONET emerged during 2017 and 2018 as a common electoral-roll management system. Before ERONET, different States used separate applications, databases, service providers, and data centres. Systems ranged from spreadsheets to State-specific database applications. ERONET standardised registration forms, database structures, electoral-roll printing, verification workflows, and reporting.
By 2020, ERONET was operating as a common database serving about 91 crore electors. The Unified National Photo Electoral Roll, or UNPER, later expanded to contain records for more than 95 crore electors across States and Union Territories.
The technological progression can be understood through several stages:
- 1997: nationwide computerisation of electoral rolls began.
- 2009: nationwide Photo Electoral Rolls were used for the Lok Sabha election.
- 2017 to 2018: ERONET created a common national electoral-roll management environment.
- Later ERONET development: Photo Similar Entries and Demographic Similar Entries became part of roll-quality checks.
- May 2025: ECINET was announced as a unified platform bringing together more than 40 existing election-related applications and websites.
- January 22, 2026: ECINET was formally launched as the ECI's unified digital election platform.
The result is that electoral-roll administration is no longer simply a collection of local electronic files. Large national databases, automated comparison tools, field applications, dashboards, citizen portals, and role-based official systems now interact throughout the voter-registration process.
ERONET, ECINET, and the Electoral-Roll Database
ERONET and ECINET perform related but different functions. ERONET is the electoral-roll management system used for registration, migration, correction, verification, deletion processing, reporting, and roll preparation. ECINET is the broader unified environment designed to bring numerous ECI services into one interface.
ERONET operates as a web-based system supporting electoral officials across India. Applications can enter the system through voter-service portals, mobile applications, BLO tools, and related interfaces. EROs and Assistant Electoral Registration Officers then process applications and field-verification information.
ERONET includes:
- a common database structure across States and Union Territories
- multilingual processing
- application tracking
- monitoring and reporting
- voter-registration processing
- field-verification records
- electoral-roll preparation
- demographic comparison reports
- photographic comparison reports
- decision-support functions for EROs
AI is used to identify Photo Similar Entries, known as PSE, and Demographic Similar Entries, known as DSE.
ECINET operates at a wider level. The platform combines voter registration, electoral-roll search, application tracking, complaints, e-EPIC services, polling information, election-official connections, candidate information, election information, and other services that previously existed across numerous separate systems.
This distinction matters when discussing the October 2026 controversy. ERONET has long contained electoral-roll management and AI-assisted PSE and DSE functions, while ECINET is the newer unified platform through which a wider range of functions and users interact.
Where Algorithms Enter the Voter Verification Process
Algorithms enter electoral-roll management primarily by comparing records and identifying entries that require human review. Public information supports the existence of AI-assisted photographic and demographic matching, but does not establish that an algorithm independently makes the final legal decision to delete an elector.
Photo Similar Entries compare voter photographs to identify records that appear visually similar. A match can indicate a possible duplicate entry, although similarity alone does not establish duplication.
Demographic Similar Entries compare combinations of information such as names, relatives' names, age, date of birth, and address. Similar demographic information can indicate duplicate registration, spelling variations, migration, data-entry errors, or different individuals who happen to have similar details.
PSE and DSE function as tools for improving electoral-roll quality. ERONET also provides decision support to electoral registration officials.
AI-supported duplicate detection received wider use during the SIR process. AI-based software was deployed across States and Union Territories undergoing SIR, including facial comparison intended to identify duplicate photographs.
The distinction between detection and decision is central.
A software match can say that two records look similar or violate a programmed consistency rule. That output does not prove that either record is invalid.
A person can share a name with another voter. Ages can be entered incorrectly. Transliteration between Indian languages and English can alter spelling. Historical rolls can contain incomplete information. Relatives can have similar names. Photographs can be old or low quality.
For this reason, an automated flag is best understood as an administrative input requiring verification rather than a legal determination about voting eligibility.
SIR Made Software-Generated Flags Much More Consequential
The Special Intensive Revision expanded the practical importance of voter-data matching because SIR requires extensive verification of existing rolls rather than relying only on routine applications for additions or corrections.
The Bihar SIR began on June 24, 2025. The exercise used house-to-house enumeration, enumeration forms, draft electoral rolls, voter verification, objections, document review, and final publication.
Phase II covered Chhattisgarh, Goa, Gujarat, Kerala, Madhya Pradesh, Rajasthan, Tamil Nadu, Uttar Pradesh, West Bengal, Andaman and Nicobar Islands, Lakshadweep, and Puducherry between October 2025 and April 2026.
On May 14, 2026, Phase III was announced for another 16 States and three Union Territories. Approximately 3.94 lakh Booth Level Officers were assigned to conduct house-to-house work involving about 36.73 crore electors, with assistance from approximately 3.42 lakh Booth Level Agents appointed by political parties.
By September 26, 2026, SIR had been completed in 20 States and Union Territories while work continued elsewhere.
By late October 2026, 17 States and Union Territories that had completed SIR were directed to undertake a Special Summary Revision with January 1, 2027 as the qualifying date. SIR continued or remained on a separate timetable in other jurisdictions.
This national scale means software matching is no longer a narrow technical issue. When millions of records are processed, even a system designed only to identify records for further checking can generate a very large administrative workload.
Mapping, Unmapped Voters, and “Logical Discrepancies”
The SIR workflow introduced another important concept, mapping current voter records against records from the previous intensive revision. Voters whose current records could be connected to their own or qualifying family records in an older roll could be treated differently from records that remained unmapped or contained inconsistencies.
During the third phase of SIR, software also generated a category described as “logical discrepancies.” Notices were issued to people who were unmapped or whose records contained logical discrepancies.
Reported discrepancy categories have included matters such as:
- differences between a current parent's name and an older record
- unusually small or large recorded age differences between parent and child
- unusual age relationships involving grandparents
- multiple people mapped to the same parent
- unusual recorded intervals between siblings
- inconsistencies between current and earlier demographic records
Such conditions can identify data requiring inspection. They do not by themselves establish that a person is illegally registered.
The scale became politically and administratively significant. Roughly 17.6 million voters across Telangana, Karnataka, Maharashtra, and Delhi were reported to have received notices connected with logical discrepancies during Phase III. That figure should be treated as a reported aggregate rather than a complete national SIR count.
The experience also showed why rule design matters. A low-level data inconsistency can become a high-impact administrative event when it produces a notice, document request, hearing requirement, or additional verification.
What Happens After Software Flags a Voter
A software flag does not legally replace the Electoral Registration Officer. Under the current framework, field verification and documentary review feed into a decision made by the officer authorised under electoral law.
The September 26, 2026 decision modified the process for electors receiving notices because they were unmapped or had logical discrepancies.
Booth Level Officers were directed to visit their homes, collect relevant documents, and upload those documents to ECINET for a decision by the ERO. Routine personal appearance at an ERO or AERO office was no longer required for these cases.
A hearing could still be ordered in exceptional circumstances, preferably online. An adult family member could also be authorised to participate when required.
District officials were also directed to create help desks or special camps for groups such as homeless people, workers, residents of night shelters, and other people who could face difficulty completing the process.
The legal framework adds safeguards beyond software design.
Section 22 of the Representation of the People Act, 1950 requires a reasonable opportunity of being heard before specified corrections or deletions are made after verification. Section 23 governs inclusion. Section 24 provides an appeal route against relevant ERO orders, followed by a further appeal to the Chief Electoral Officer under the statutory framework.
This creates a hierarchy:
software analysis → field verification → documents and records → ERO determination → statutory appeal or judicial review where applicable
That hierarchy is important because an automated matching system can affect which records receive scrutiny without itself possessing statutory authority to determine voter eligibility.
The Supreme Court’s May 2026 SIR Ruling
The Supreme Court's May 27, 2026 judgment established the main legal framework within which SIR and software-assisted verification now operate. The Court upheld the ECI's authority to undertake the Bihar Special Intensive Revision under Article 324 of the Constitution and Section 21(3) of the Representation of the People Act, 1950.
The litigation examined several broad issues, including the Commission's power to order an intensive revision, proportionality of the process, compliance with electoral law, and the ECI's authority to examine citizenship for electoral enrolment.
The Court upheld the SIR methodology while retaining limits on what the ECI can determine.
The judgment said the ECI can conduct a limited inquiry into citizenship where citizenship is relevant to eligibility for inclusion in the electoral roll. The Commission's electoral decision does not itself constitute a final legal determination that a person is or is not an Indian citizen.
Where citizenship requires formal determination, the matter must go to the competent authority under citizenship law.
The Court also preserved judicial review for people who contend that they were wrongly removed on grounds such as absence, death, migration, or duplication.
The judgment therefore strengthens the legal authority for intensive voter verification while maintaining a distinction between electoral eligibility decisions and final citizenship determinations.
It does not resolve a different issue now receiving attention, namely how much information about the software generating verification flags should be publicly disclosed.
The October 2026 RTI Refusal
The current transparency debate stems from a Right to Information application filed on July 27, 2026 seeking details about the technology used to identify errors, anomalies, and inconsistencies during SIR.
The applicant sought information about the software, application, algorithm, or AI platform used during the exercise. The questions also covered whether voter records were automatically marked for verification, correction, or deletion, what happened after a record was identified, and whether AI, machine learning, predictive analytics, or automated decision-support technology was involved.
The ECI replied on October 16, 2026.
The Commission said no new software had been developed or procured specifically for SIR.
It also said EROs determine inclusion or exclusion on the basis of facts and records.
The ECI did not disclose the requested software name and version. It relied on the RTI Act exemption relating to commercial confidence, trade secrets, and intellectual property.
For some questions concerning automated systems and workflow, the ECI responded that the request sought clarification rather than “information” falling within Section 2(f) of the RTI Act.
Section 8(1)(d) of the RTI Act provides protection for commercial confidence, trade secrets, and intellectual property where disclosure would harm the competitive position of a third party. The statutory wording also contains a larger-public-interest qualification.
The October response therefore establishes several facts but leaves several technical matters unresolved.
It confirms that existing technology rather than newly procured SIR-specific software was used. It confirms that the ERO remains the stated human decision-maker. It does not publicly identify the precise software version or give a detailed description of every automated rule operating within the SIR workflow.
Transparency Does Not Require Publishing Every Line of Source Code
The policy choice is not limited to either complete secrecy or unrestricted publication of source code. Meaningful transparency can exist at several levels without exposing credentials, security mechanisms, personal voter information, or technical material that could create security risks.
For an electoral-roll decision-support system, useful public documentation can include:
- the name and version of the production system
- the purpose of each automated check
- categories of data examined
- definitions of each discrepancy category
- whether each rule is deterministic, statistical, or AI-based
- whether a flag can trigger action automatically
- which officials can override or close a flag
- what verification must occur before deletion
- change history for rules affecting voter scrutiny
- aggregate numbers of voters flagged by category
- aggregate outcomes after human verification
- documented access-control responsibilities
- audit methodology
- correction and appeal pathways
Such information would allow the public to distinguish a simple consistency check from an AI model, a probable duplicate from a confirmed duplicate, and a software-generated notice from an ERO order.
Source code can raise separate intellectual-property and cybersecurity issues. Process documentation, audit results, rule descriptions, version history, and aggregate accuracy information are different forms of disclosure.
That distinction is particularly relevant because the ECI has publicly confirmed that AI is used for PSE and DSE functions within ERONET while declining to disclose certain additional technical details requested through RTI.
Centralisation and Access Control Became a Separate Issue
Software transparency also concerns who controls electoral data and who can exercise functions inside the system. Electoral law gives the ERO responsibility for preparing and revising the roll, so software permissions must allow that statutory role to operate.
Concerns emerged during September 2026 over whether some ECINET design choices could restrict the practical access or discretion available to field election officers.
The ECI stated that field officials have role-based ECINET access according to their statutory powers.
On September 26, 2026, the full Commission created a review committee headed by a Senior Deputy Election Commissioner and including an independent expert from an IIT or IIIT. The committee was asked to check ECINET against the governing Acts and Rules.
The Commission also said additional flexibility requested by field officials would be made operational and that future IT initiatives would go through an internal officer committee before Commission approval.
The September announcement did not specify a public completion date for the ECINET review or state that the resulting report would necessarily be released publicly.
A later examination of publicly served ECINET code identified access-control features that were considered worthy of further scrutiny. The examination could not establish from public code alone whether the reported restrictions had actually been applied at the server level or how frequently any such controls were used.
That limitation matters. A software feature's existence, its technical availability, its administrative authorisation, and its actual use are four separate facts.
Cybersecurity and Algorithmic Transparency Are Different Problems
Cybersecurity concerns whether unauthorised people can access or alter a system. Algorithmic transparency concerns how authorised software processes information and influences administrative decisions. Electoral technology needs both forms of accountability, but disclosure appropriate for one area can be unsuitable for the other.
In October 2026, a security researcher was reported to have submitted ECINET-related issues to India's Computer Emergency Response Team, CERT-In, in July.
CERT-In later communicated that one reported issue had been fixed while other reported matters remained under work at that point.
Security concerns should not be confused with proof that electoral-roll records were altered or that SIR results were manipulated. The publicly available information does not establish such a finding.
The episode does show why election technology requires several independent controls:
technical security testing, strict role permissions, change logging, voter-level legal safeguards, administrative audits, and clear public documentation of how software-generated flags affect official workflows.
Publishing sensitive security material can itself create risk. Publishing high-level audit outcomes, remedial status, access-control governance, and non-sensitive decision rules presents a different trade-off.
What the ECI Has Disclosed and What Remains Unclear
The public record as of October 30, 2026 provides considerably more information about India's voter-management technology than the RTI dispute alone might suggest, but important operational details remain unspecified.
The ECI has publicly stated that:
- ERONET is used for national electoral-roll management.
- UNPER contains records for more than 95 crore electors.
- ERONET supports field verification and ERO decision support.
- AI is used for Photo Similar Entries and Demographic Similar Entries.
- ECINET integrates more than 40 earlier election applications and services.
- SIR uses software-supported mapping and discrepancy processing.
- EROs retain legal authority over voter inclusion and exclusion.
- no new software was developed or procured specifically for SIR.
- voters identified as unmapped or logically inconsistent now generally receive document collection through BLO home visits.
- ECINET is undergoing a compliance review involving an independent IIT or IIIT expert.
Information that has not been publicly established in sufficient detail includes:
- the exact software name and version referred to in the July 2026 RTI questions
- complete technical specifications for SIR anomaly detection
- all production matching thresholds
- complete definitions for every logical-discrepancy rule across every State
- validation results for individual automated checks
- false-positive rates for PSE, DSE, mapping, or logical-discrepancy flags
- the full relationship between ERONET modules and current ECINET roll-management modules
- a public version history showing when individual rules were introduced or changed
- complete aggregate figures showing how many software flags were confirmed or rejected after field verification
- the completed report of the September 2026 ECINET review, based on information available as of October 30, 2026
These gaps do not establish wrongdoing. They define the boundary between information currently available to the public and information that has not been publicly specified in sufficient detail.
Why Human Review Remains the Central Safeguard
The central issue in algorithmic voter management is not whether software is used. At India's electoral scale, software has been part of voter-roll administration for decades. The more important issue is what authority software outputs carry and what safeguards exist before an elector's legal status changes.
A well-defined decision-support model keeps separate stages separate.
A demographic similarity is not a duplicate.
A photograph similarity is not proof that two registrations belong to the same person.
Failure to map against an older roll does not by itself establish ineligibility.
A logical inconsistency is not a deletion order.
A notice is not a final determination.
The stated process places the final statutory decision with the ERO. The Representation of the People Act requires verification and provides hearing and appeal protections. Judicial review is also available in disputes involving wrongful deletion.
The quality of algorithmic voter management therefore depends on both technical accuracy and administrative design. False matches must be correctable. Officers must retain lawful discretion. Voters must know why additional verification is required. Decisions must be traceable to the responsible authority. Appeal routes must remain usable.
The Current Position as of October 30, 2026
India now operates one of the world's largest software-supported voter-management systems. Electoral rolls that began nationwide computerisation in 1997 are now managed through national databases, automated similarity checks, AI-supported duplicate detection, digital field workflows, and the ECINET unified platform.
The legal position is clearer than the technical disclosure position.
The Supreme Court has upheld the ECI's authority to conduct SIR and recognised a limited voter-eligibility inquiry connected with citizenship, subject to legal boundaries and review. Electoral Registration Officers remain the statutory decision-makers for roll inclusion and deletion.
The technical position is partly documented. The ECI openly confirms AI use for PSE and DSE, centralised digital form processing, analytical reports, and decision-support functions. SIR has also introduced large-scale mapping and logical-discrepancy processing.
The transparency dispute remains unresolved. The October 16 RTI response did not identify the software name and version requested by the applicant or provide a complete account of the automated logic involved. The ECI relied on RTI protections concerning commercial confidence, trade secrets, and intellectual property, while stating that EROs make the actual electoral decisions.
At the same time, the Commission has ordered a separate ECINET compliance review involving an independent technical expert. No completed public report from that review was identified in the material available up to October 30, 2026.
The central policy issue is therefore narrower than the broader political debate. India already uses automated and AI-assisted methods to identify voter records requiring attention. The unresolved question is how much non-sensitive information about those systems, their rules, versions, error patterns, access controls, and audit results should be made public while preserving voter privacy, intellectual property, and cybersecurity.
Algorithmic voter management has become a major part of India’s electoral-roll administration, moving from basic computerisation to centralized databases, automated matching, AI-assisted similarity checks, and software-supported verification through systems such as ERONET and ECINET. These tools can help officials identify duplicate records, demographic inconsistencies, mapping problems, and other entries that require further examination, but a software-generated flag does not itself determine whether a voter should remain on or be removed from the electoral roll.
The Special Intensive Revision has increased the importance of these systems because automated checks can now lead to notices, document verification, field visits, and additional scrutiny for large numbers of electors. Electoral Registration Officers remain responsible for final statutory decisions, with hearing, appeal, and judicial-review protections forming part of the legal process.
The October 2026 RTI refusal has shifted attention from whether electoral software exists to how transparent its operation should be.
The ECI has publicly confirmed the use of AI-assisted voter-record matching and decision-support systems, while declining to disclose some software details, including the requested name and version, citing protections related to commercial confidence, trade secrets, and intellectual property.
The main unresolved issue is therefore accountability around automated voter management. Clear documentation of discrepancy rules, software versions, human-review requirements, error rates, access controls, audit findings, and correction procedures could help distinguish automated screening from final legal decision-making.
As India expands digital election administration, the credibility of algorithm-assisted voter management will depend not only on technical performance but also on transparent procedures, human oversight, voter safeguards, and clear responsibility for every decision affecting the electoral roll.
Algorithmic Voter Management in India: FAQs
Algorithmic voter management refers to the use of digital databases, automated matching systems, analytical rules, and AI-assisted tools to help maintain electoral rolls. These systems can identify possible duplicate records, demographic inconsistencies, photo similarities, mapping problems, and other entries that may require verification.
ERONET is an electoral-roll management system used for voter registration, corrections, migration, verification, deletion processing, field verification, reporting, and preparation of electoral rolls. It also supports Photo Similar Entries and Demographic Similar Entries checks.
ERONET primarily focuses on electoral-roll management and voter-record processing. ECINET is a broader unified election platform that brings together multiple election-related services, applications, voter tools, administrative functions, and digital workflows under one system.
AI-assisted systems can compare voter photographs and demographic information. Photo Similar Entries identify records with similar photographs, while Demographic Similar Entries compare information such as names, relatives' names, age, date of birth, and addresses. A similarity flag requires further verification and does not automatically prove duplication.
The stated electoral process does not give software the final legal authority to delete a voter. Automated systems can flag a record for examination, but Electoral Registration Officers are responsible for decisions involving inclusion, correction, or deletion after considering records, verification, and applicable legal requirements.
Logical discrepancies are inconsistencies detected in voter records that may require further examination. Examples can include unusual age differences between family members, differences in names across older and current records, multiple people mapped to the same parent, or inconsistencies involving demographic information.
A flagged record can be sent for additional verification. Booth Level Officers may collect information or documents, field verification can take place, and the relevant material can be reviewed by the Electoral Registration Officer. A software flag itself is not a final determination of voter eligibility.
In its October 2026 RTI response, the Election Commission declined to disclose certain requested software details, including the requested name and version. It relied on RTI provisions relating to commercial confidence, trade secrets, and intellectual property. The Commission also stated that no new software had been developed or procured specifically for SIR.
Transparency helps voters and election officials understand why a record was flagged, what data was considered, what type of automated rule was applied, and how a final decision was reached. Information about discrepancy categories, review procedures, software changes, audit processes, and error rates can improve accountability without requiring disclosure of sensitive security information.
Voter safeguards include human review by Electoral Registration Officers, field verification, documentary examination, opportunities to be heard in relevant cases, statutory appeal mechanisms, and judicial review where applicable. These safeguards are important because an automated similarity or discrepancy flag does not by itself establish that a voter is ineligible.