Electoral Roll SIR: AI Anomaly Detection Raises Due-Process Questions

Why in News?

The Hindu reported on July 26, 2026 that an artificial-intelligence tool was flagging logical discrepancies after Booth Level Officers digitised enumeration forms during the Special Intensive Revision (SIR) in Bengaluru. The Greater Bengaluru Authority disclosed at least 6.41 lakh flags within its jurisdiction, but the category describes potential anomalies requiring verification, not 6.41 lakh proven duplicate, ineligible or invalid voters.

Separate reports from Karnataka described enrolled voters being marked “shifted” despite continuing at their recorded addresses, while Haryana said 33,84,568 names went unrecorded during its first enumeration phase for reasons including relocation, death, duplication, absence and non-submission. Read together, the developments make accuracy, notice, hearing, correction, appeal and meaningful human review central to electoral-roll integrity.

• Within the GBA jurisdiction, at least 6.41 lakh logical discrepancies had been flagged when only 38.81% of enumeration forms had been digitised as of 4 p.m. on July 25; the total included 2.66 lakh flags in Bengaluru Urban.
• The tool also uses Demographic Similar Entries (DSE) to identify records sharing details such as names or fathers’ names, but a similarity flag by itself doesn’t establish that two records belong to one person.
• Bengaluru’s electorate exceeds 1.03 crore, and more than 15.83 lakh electors were categorised as unmapped or with no mapping because their forms lacked traceable 2002-roll details.
• In Haryana, 20,629 BLOs distributed forms to more than 2 crore electors and more than 1 crore forms were digitised; 33,84,568 names remained unrecorded at the end of phase I.
• Haryana scheduled its draft roll for July 31, claims and objections from July 31 to August 30, and final roll for October 3, making the draft-and-remedy stage decisive.

The development matters in the context of:

• The constitutional aim is inclusion with integrity: every eligible citizen should be enrolled, while ineligible or duplicate entries should be removed only through lawful verification.
• At population scale, automated screening can narrow a review queue, but false positives can transfer the cost of correcting official data to genuine electors.
• The issue is a fresh administrative trigger, distinct from the earlier Supreme Court ruling on SIR deletions: the present question is how algorithmic flags are converted, or not converted, into official roll decisions.

UPSC Relevance

Prelims Relevance

• Article 324 vests the superintendence, direction and control of elections in the Election Commission of India.
• Article 326 provides for elections to the Lok Sabha and State Legislative Assemblies on the basis of adult suffrage, subject to constitutional and statutory disqualifications.
• The Representation of the People Act, 1950 governs qualifications of electors and preparation, revision, correction and inclusion in electoral rolls; it should not be confused with the 1951 Act’s main focus on conduct of elections and electoral offences.
• Section 17 bars registration in more than one constituency, while Section 18 bars registration more than once in the same constituency.
• Section 19 links enrolment to the age threshold and ordinary residence; Section 20 explains ordinary residence.
• Section 21 concerns preparation and revision of rolls, while Section 22 permits correction, transposition or deletion by the ERO after inquiry and proper verification.
• Section 23 provides for inclusion of a name on application after verification, and Section 24 provides the statutory appeal framework.
• Form 6 is used to seek inclusion as a new elector; Form 7 is used for objection or deletion; Form 8 covers shifting of residence, correction, replacement of EPIC and marking of disability.
• Under Rule 21A of the Registration of Electors Rules, 1960, a person proposed for deletion during revision should receive a reasonable opportunity to show cause.
• A BLO assists with household-level verification, but the Electoral Registration Officer is the statutory decision-maker for roll entries.

Mains Relevance

GS Paper 2

• Electoral governance: balancing universal inclusion, roll purity, federal field administration and the ECI’s constitutional responsibility.
• Procedural fairness: notice, disclosure of the case, opportunity to respond, reasoned orders and effective appeals before an adverse entry becomes final.
• Institutional accountability: allocating responsibility among the ECI, Chief Electoral Officer, ERO, AERO, BLO, vendors and local authorities.

GS Paper 3

• Technology governance: assessing data quality, false-positive rates, explainability, audit trails, cybersecurity and human oversight in high-impact public decisions.
• Digital public systems: ensuring that scale and speed don’t substitute for verified facts when civic participation is at stake.

Essay

• Efficiency and justice: a fast administrative system isn’t a fair system unless citizens can understand and challenge consequential errors.
• Technology and democracy: algorithms may assist public officials, but legitimate authority and responsibility must remain human and reviewable.
• Invisible exclusions: small data errors can become large democratic harms when replicated across mass databases.

Background and Context

What SIR Is Designed to Do

A Special Intensive Revision is an ECI-directed, field-heavy revision intended to rebuild or closely verify the electoral roll against the statutory eligibility conditions.

• The stated objective is symmetrical: no eligible citizen left out and no ineligible person included, rather than deletion as the exercise’s default purpose.
• During enumeration, BLOs conduct house-to-house visits, distribute and collect forms, and support verification; electors may also use the online channel where available.
• The draft roll isn’t the final roll. Publication opens a structured period in which omissions, wrong particulars and objections can be identified and adjudicated.
• The legal foundation rests on Article 324, Section 21 and other applicable provisions of the 1950 Act; an earlier overview of the judicial position is available in Supreme Court upholds ECI’s SIR.

How the Bengaluru Anomaly Tool Works

The reported system acts as a screening layer after a BLO digitises and submits an enumeration form; it flags records for scrutiny rather than conclusively deciding voter eligibility.

• The disclosed 6.41 lakh refers to logical discrepancies in the wider GBA jurisdiction at an interim stage, while 2.66 lakh were attributed to Bengaluru Urban.
• Officials described DSE matching, under which identical or similar demographic details across records, including names and parental names, may be surfaced as possible duplicates.
• Ordinary Indian naming patterns can create ambiguity: initial expansion, spelling variation, transliteration, reordered names and family members sharing names can produce similarity without duplication.
• Officials also reported implausible flags, including a system inference that two children linked to one parent were born less than nine months apart; that example illustrates why a rule can misread source data or relationships.
• The GBA report didn’t disclose the anomaly taxonomy, and the Statewide progress report reportedly didn’t show this category, limiting public understanding of what the aggregate actually measures.
• Officials said a data-scrubbing exercise would remove wrongly flagged cases before a mechanical review. Scrubbing quality and independent human judgment matter because automation can reproduce the error embedded in its inputs.

Why 'Shifted' Can Be a Data Error

The label shifted should reflect ordinary-residence facts, not merely a mismatch between historical lists, polling-station boundaries or field assignments.

• Voters in Yeshwanthpur and other Bengaluru areas told The Hindu that one household member was marked shifted while other members at the same address received unmarked forms.
• Some BLOs reportedly connected the label to differences between 2002 and 2025 addresses; other field explanations pointed to delimitation and transfer of streets or apartment blocks to a new polling station.
• A voter can remain at the same dwelling while the polling station, part number or BLO assignment changes. Administrative movement of a record isn’t proof that the person ceased to be ordinarily resident.
• The reported advice to submit Form 8 can repair a shift or incorrect entry, but requiring a correctly recorded elector to correct an error they didn’t create raises an access and burden question.
• Karnataka’s Chief Electoral Officer said a person is marked shifted only when unavailable at the address and said no such complaints had reached him. The gap between field reports and headquarters visibility points to a grievance-escalation problem.

What the Haryana Numbers Establish

Haryana’s figures describe names unrecorded at the enumeration stage; the totals shouldn’t be presented as final deletions before the draft, claims, hearings and final roll.

• The 33,84,568 total comprised 13,75,278 permanently shifted, 7,66,205 reported deceased, 2,04,916 already registered, 9,36,646 absent and 1,01,523 who didn’t submit forms for other reasons.
• An absent elector isn’t automatically dead, shifted or ineligible. Temporary work, study, travel, illness, locked premises and failed contact can all require another verification attempt.
• A record categorised as already registered indicates a possible duplication issue, but Sections 17 and 18 require identity-level verification before deciding which entry, if any, should be struck off.
• The CEO said an omitted elector may use Form 6 after the July 31 draft, and no name would be deleted without an opportunity of hearing.
• The Haryana schedule allowed claims and objections until August 30. The CEO also described appeal windows of 15 days to the District Magistrate and a further 30-day appeal to the Chief Electoral Officer.
• Political parties had appointed 37,055 Booth Level Agents to assist. BLAs can support scrutiny and outreach, but the ERO retains responsibility for lawful individual decisions.

The Due-Process Chain

An algorithmically generated queue becomes constitutionally significant only when an official treats it as a basis for notice, burden-shifting, correction or deletion.

• First, verify the facts: Section 22 authorises the ERO to correct, transpose or delete after inquiry and proper verification; a machine score can’t replace evidence of death, change of ordinary residence or ineligibility.
• Second, give intelligible notice: the elector should know the proposed action, the specific mismatch, the underlying record and the deadline for response, not receive an unexplained anomaly label.
• Third, hear the elector: Rule 21A reflects the principle of audi alteram partem, or hearing the other side, through a reasonable opportunity to show cause before deletion.
• Fourth, record reasons: a reasoned ERO order shows whether the officer independently assessed evidence and makes an appeal meaningful.
• Fifth, preserve review: claims, corrections and appeals need accessible online and offline channels, acknowledgement numbers, status tracking and remedies for people with disability, migration or weak connectivity.
• This chain matches the wider lesson from the July 18 SIR decision: an electoral-roll determination must remain confined to electoral law and can’t silently become a wider declaration of civic status.

Algorithmic Accountability in Election Administration

High-impact public technology needs measurable safeguards because errors can be difficult for an individual voter to detect before election day.

• Purpose limitation: define whether each model or rule detects duplication, demographic inconsistency, address mismatch or missing historical mapping; don’t collapse unlike risks into one number.
• Quality metrics: publish category-wise flag counts, sampling results, false-positive rates and correction outcomes without exposing personal electoral data.
• Traceability: retain the source record, rule or model version, time of flag, reviewer action and final reason so an audit can reconstruct the decision path.
• Human-in-command: trained officials should test the flag against field evidence and law; rubber-stamping a dashboard is nominal, not meaningful, review.
• Vendor accountability: procurement contracts should permit technical audit, incident reporting and correction, while the ECI and ERO remain answerable for public decisions.
• Interoperability caution: linked databases can improve verification, but name, address and life-event records carry different error rates. The Civil Registration System data lesson shows why coverage, lag and ground verification still matter.

Democratic Significance

Electoral-roll quality is not a narrow database-management issue; it determines whether the constitutional promise of adult suffrage can be exercised in practice.

• False inclusion can enable duplication or weaken confidence, while false exclusion directly denies an otherwise eligible citizen the chance to vote.
• The burdens aren’t evenly distributed: migrant workers, tenants, women changing residence after marriage, students, elderly electors and people without easy document access may face more mismatches.
• Aggregate accuracy can hide concentrated harm. Even a low false-positive rate can affect many people when the system processes crores of records.
• Public trust depends on both roll purity and procedural legitimacy. A correct outcome reached through an opaque process may still be hard to defend, reproduce or distinguish from an accidental result during independent review.
• The safest institutional framing is that technology supports administrative attention; eligibility, correction and deletion remain legal judgments grounded in verified facts, transparent procedures and an officer’s recorded responsibility for the final action.

Way Forward

Separate Flags from Findings

• Label every automated output as a review flag, prohibit direct deletion from a score, and require an ERO-authorised finding based on documented evidence.
• Use category-specific thresholds and manual sampling so similar names, address changes and historical mapping gaps aren’t treated as one class of risk.

Build Notice Around the Actual Mismatch

• Send a plain-language notice stating the exact field or record conflict, proposed consequence, response route, documents that may help and the hearing timeline.
• Provide notices through multiple channels without making a mobile number mandatory, and preserve doorstep and designated-centre access.

Strengthen Human Review

• Give BLOs and ERO staff written decision rules, training on delimitation and ordinary residence, escalation support and enough time to investigate exceptional cases.
• Audit whether reviewers merely accept the flag or record an independent reason; accountability requires checking the quality of human review, not only the presence of a human click.

Publish Safe Performance Evidence

• Release district- and category-level counts for flags, notices, accepted corrections, rejected flags, hearings and restored names, with privacy-protective aggregation.
• Commission periodic technical and process audits covering data provenance, false positives, access controls, model changes and grievance outcomes.

Make Remedies Usable Before Finalisation

• Run public search-and-check campaigns after the draft roll, provide assisted Form 6 and Form 8 filing, issue acknowledgements and disclose disposal reasons.
• Track restoration times and appeal pendency, and prioritise cases close to final publication so an elector isn’t left with a paper remedy after the voting opportunity has passed.

Conclusion

The Bengaluru and Haryana developments don’t prove that electoral-roll technology is inherently exclusionary. They show that scale magnifies both efficiency and error: millions of records can be screened quickly, but ambiguous labels can also be propagated before citizens know that their status is at risk.

The defensible model is simple: let technology identify records needing attention, but keep verification, notice, hearing, reasons and appeal between a flag and an adverse roll decision. Electoral integrity is protected when inclusion and roll purity are pursued together under accountable human authority.

UPSC Practice Questions

Prelims MCQ 1

With reference to electoral rolls under the Representation of the People Act, 1950, consider the following statements:

1. Section 17 bars a person from being registered in more than one constituency.
2. Section 22 empowers the Electoral Registration Officer to correct, transpose or delete entries after inquiry and proper verification.
3. Form 8 is the prescribed application for a person seeking first-time inclusion as a new elector.

How many of the above statements are correct?

(a) Only one (b) Only two (c) All three (d) None

Answer: (b) Only two

Explanation:

Statements 1 and 2 are correct. Form 6 is used for inclusion as a new elector. Form 8 covers shifting, correction, EPIC replacement and marking of disability.

Prelims MCQ 2

Which safeguard is most directly associated with Rule 21A of the Registration of Electors Rules, 1960?

(a) Automatic deletion of every demographically similar entry (b) Parliamentary approval before revision of a State electoral roll (c) A reasonable opportunity to show cause before proposed deletion (d) Transfer of the ERO’s decision-making power to a software vendor

Answer: (c) A reasonable opportunity to show cause before proposed deletion

Explanation:

Rule 21A requires an effort to give the person concerned a reasonable opportunity to show cause. An algorithmic similarity flag isn’t a substitute for this procedural protection.

UPSC Mains Questions

1. Large-scale anomaly detection can improve electoral-roll administration, but an algorithmic flag cannot become an adverse legal finding by default. Examine the safeguards needed to reconcile roll purity with universal inclusion, natural justice and effective appellate remedies. (250 words)
2. The quality of human oversight matters more than the mere presence of a human reviewer in high-impact digital governance. Discuss with reference to electoral-roll revision, data quality, explainability, audit trails and institutional accountability. (250 words)

Sources: Election Commission of India and The Hindu.

Frequently Asked Questions