Author: Ojas Sharma from Integrated Law Course, Faculty of Law, University of Delhi
& Co Author: Vishu Mittal from Integrated Law Course, Faculty of Law, University of Delhi
I. Introduction: From Bureaucracy to Softwarecracy
The promise of a democracy begins with a system of free and fair elections, with its core tenet being the electoral roll. Votes constitute not mere data points; they are the very system that sustains a democratic initiative. The importance of free and fair elections has been reiterated by the Supreme Court in cases as early as Indira Gandhi v. Raj Narain. Thus, what becomes evident is that such a necessity cannot be dispensed with, and it certainly cannot be subjected to arbitrary
systems that prejudice the interests of the masses in favor of botched efficiency.
The electoral system has undergone a significant evolution in the modern age. From the
introduction of Electronic Voting Machines (EVMs) to digital data logging and online voter-ID registration, we have witnessed new avenues to the oldest system of political organization. However, a stark problem arises when the very essence of this system is sought to be disbanded:
humans. This problem has manifested when the ECI sought to implement an AI software in its Special Intensive Revision (SIR) procedure in West Bengal. The accuracy level of this AI system was explained by experts to be around 60% to 90%.1 While mathematically a 90% accuracy soundspromising, in reality it rings hollow for a state that accounts for over 7% of the population of the country. The 10% error margin encapsulated approximately 76 lakh adults whose voting rights ran the risk of being suspended even before the implementation began. What unfolded was far worse than what anyone could have imagined. Over 90 lakh voters were excluded by the SIR in West Bengal.2 The AI did not just erase votes; it also disentitled those 90 lakh people from having access to all government schemes they previously enjoyed. It erased identities and it erased sustenance. What further confounds the electoral result is that certain demographic groups were subjected to “additional scrutiny” by this AI and faced disproportionate deletions in their constituencies.3
The manipulation of the electoral roll is not new; it has stained politics since the days of Elbridge Gerry.4 Yet, while voter suppression was once an arduous, highly visible task, AI technology has made it scalable and chillingly anonymous. What we see today is a successful experiment in bending the electoral roll to favor a state that demands compliance and control. When the state replaces human labor with technological “efficiency,” it risks collapsing into a “softwarecracy.” A republic is a constant negotiation between a state seeking a legible, predictable populace and citizens seeking the realization of constitutional ideals like sovereignty and
representation. A truly representative state is bound by a democratic oath to its governed. But when the limbs of bureaucracy—once anchored by human accountability—are shifted to artificial systems shielded by an illusion of mathematical infallibility, we must ask: who is left to be held responsible?
II. The Anatomy of an Algorithmic Purge: The Bengal SIR Debacle
The Special Intensive Revision, framed as a necessary administrative cleanup, reveals heavy reliance on opaque algorithmic deduplication processes. This automated verification software has shifted the burden to vulnerable citizens by flagging their profile as “doubtful.” One must ask, what fulfills the criterion of the said doubt? It signals towards “logical discrepancies,” which include voters with mismatched name spellings or a single individual linked as a parent to more than six voters, or even where the age difference between a voter and grandparent is less than 40 years.5These data-entry errors made by the officials pushed voters into an “under adjudication”
category, subsequently rivaling nearly 12% of the voter base from 7.66 crore to 6.75 crore6,6 which is one of the largest mass voter deletions in recent democratic history executed by ECI. Despite production of valid documents asserting their citizenship and right to vote, the ECI hearings made no difference.
This assault on the voter’s right is not merely a consequence of ineffective hearings before the Election Commission of India (ECI), but rather a product of algorithmic governance deployed without adequate safeguards. According to empirical findings by the Sabar Institute, human review ultimately retained 3.25 million citizens on West Bengal’s electoral rolls while deleting 2.71 million during the automated verification sweep. An estimated 54% false-positive rate reflects a catastrophic inaccuracy in the automated flagging system, raising severe questions that are not just technical but constitutional. If human review reverses 99.3% of algorithmic decisions, the algorithm is not performing verification. It is performing exclusion, with the burden placed on the
citizen to undo it.The consequences and results of Bengal’s SIR did not remain contained to the elections but birthed a cascading penalty. The deleted rolls were drawn from daily wage workers, women in the informal economy, and members of specific religions. For these individuals, ration cards or government schemes are not merely convenient. It is security and a lifelong benefit meant to sustain
generations. The harm suffered is not merely administrative but also existential.
III. Core Constitutional and Philosophical Argument
A. Doctrine of Irreversible Constitutionality
The case of Bengal illustrates a distinct question on the horizon: when automation meets sovereignty, who should win? The matter of the importance of elections has never been up to debate in Indian Democracy. While the election commission of India holds wide powers to conduct elections under Article 324 of the constitution, this power is subject to reasonable safeguards.
The hon’ble Supreme Court in Shailesh Manubhai Parmar v. ECI held that in a democracy the purity of elections is categorically imperative.This purity is ensured through an independent and impartial election commission. The SC in Anoop Baranwal went so far as to say that this independence forms a part of the basic structure of the constitution, as it is what gives rise to substantial democracy.However, in the same judgment, it was also held that democracy is based
on adult franchise, and the right of adults to take part in the election process could only be restricted by a valid law that does not offend constitutional provisions.
Justice Krishna Iyer in Mohinder Gill v. Chief Election Commissioner famously stated that ECI’s powers operate within two clear boundaries; it cannot act contrary to an express law, and its actions must conform to the principles of natural justice and fairness. The importance of a free and fair electoral process thus cannot be understated. A physical manifestation of that fairness operates
through the electoral roll. In the Radhey Shyam Dani case, a single clerical error led the Constitutional Bench of the Supreme Court to set aside an entire election on the grounds that an imperfect electoral roll where citizens are denied the opportunity for scrutiny and revision is fundamentally invalid. The Apex Court has consistently upheld that while an unrevised electoral roll does not invalidate an election,18 a faulty or erroneous one defeats the very principles on which substantial democracy is built. If legitimate citizens cannot elect their representatives, it extinguishes the spirit of a democratic project in its totality. Further, it is apparent that Article 326 is a time-sensitive democratic trigger that operates on the day of elections. It cannot be actualized ex post facto. Administrative law cures rely on retroactive
application; however, within electoral jurisprudence, a delayed remedy is equivalent to a denied right. If the “purity of sufficiency” of a democracy is realized at the polling booth, the duty of the election commission to ensure a free, fair, and immediate election becomes a constitutional mandate. If Article 329(b)20 and Section 8021 of Representation of People Act, preclude any interference in an ongoing electoral process by even the Apex Court, making it a time-bound, uninterrupted pipeline, then an administrative purge that disentitles valid citizens from voting creates an infructuous electoral victory. A 99.3% reversal rate might enable that citizen to secure voting rights in the future, but their participation in that democratic cycle becomes effectively null.
This then becomes an irreparable injury under both equity and the jurisprudence established in Dalpat Kumar v. Prahlad Singh.
B. Algorithmic Suspension of Democracy
Legitimacy and citizenship are not mere political terms; they have an undeniable social dimension of belongingness. When that belongingness is denied, it leads to societal alienation, the breakdown of democratic institutions, and the creation of what Hannah Arendt refers to as “superfluous” people. All these conditions function as fodder for the origin of totalitarian regimes. In Arendt’s terminology, then, citizenship would be what confers a ‘right to rights. ‘ It thus becomes a precondition to a valid democracy; treating citizenship as a revocable license leads to a collapse of the very ideal on which a democracy is constructed. The right to vote is a constitutional right of any citizen; a suspension of this right effectively serves as a dilution of citizenship. SIR, as a mechanism needs to be understood as distinct from any previous electoral
roll revisions that India has undergone, the present mechanism flips the burden of responsibility of being on the electoral roll from the state onto the citizen. Shifting of this burden introduces into the matrix of election preparation costs, which have significant adverse implications for voters belonging to a lower socio-economic stratum. Low-earning voters are unlikely to seek adjudication against infringement of their right to vote, as the litigation costs may be greater than their present economic capabilities. The current SIR process watered down the mechanism for exclusion by not defining any criterion for exclusion specifically; individuals were flagged arbitrarily at the discretion of authorities. This makes evident the fact that even before the incorporation of an automated agent, the exclusion of citizenship can be effectively made a routine
exercise; the problem, however, becomes even more stark when AI is introduced into the matrix and the human element is removed.
The ECI attempted to justify the use of AI software specifically in Bengal on several grounds; it was even stated that while AI will aid verification, it would still be the responsibility of BLOs to conduct door-to-door verifications to ensure strict accountability. However, that accountability served no purpose when the AI software flagged over 90 lakh voters and put them in an under adjudication category, suspending their right to vote. In such a situation, it became a physical
impossibility for BLOs to be able to verify all flagged names in time for the election. The crushing and inhumane workload of SIR was amplified by problems due to the incorporation of AI and left a trail of suicides by the foot soldiers of electoral democracy.What the assembly elections of Bengal thus revealed was a case of softwarecracy that shifted the burden of proof of legitimacy and citizenship onto the citizens. The AI model—by imposing the ‘under adjudication’ status—created a presumption of illegitimacy, a clear procedural and constitutional fallacy.
C. Systemic Bi-Modal Errors
The deployment of an AI model is often justified on the grounds of neutrality. It is posited that an algorithm has no bias, has no incentive, and works in an objective manner focusing on efficiency and productivity. Nothing could be further from the truth. The growing integration of AI systems in public administration has raised serious concerns among theorists. An emerging corpus of literature suggests that deployment of AI is disruptive to foundations of democratic constitutionalism. It leads to a situation where governance and administration are structured in
what is purported to be a depoliticized domain but in reality functions in a way that is biased by data structures on which it is built. Biased data sets create discrimination, which is seemingly normalized under the guise of algorithmic neutrality.36 The arbitrary nature of the AI data set deployed in Bengal becomes evident from analysis of experience of exclusion. A data system that manifestly targeted certain groups and constituencies over others cannot be termed objective.
A manifestly capricious and unreasoned administrative action is liable to be struck down within the established jurisprudence of Article 1437, and subject to review by courts under writ jurisdiction. But the situation becomes more convoluted when one places it in the context of a black-box system. The algorithmic dimension of the AI deployed during the SIR was never revealed to the public; this made it un-auditable against any form of discrimination. All the aggrieved could do was approach the judiciary ex post facto their exclusion from the electoral roll, but neither the judiciary nor the aggrieved could ever discover why this exclusion was made. Justice P.N. Bhagwati has laid down that arbitrariness is the antithesis of equality; wherever it exists in state actions, Article 14 springs to life. An AI model trained on skewed data does not shed its bias by the virtue of automation; it launders that bias into an appearance of objectivity. The SIR mechanism deployed in Bengal is a textbook instance of this laundering: a system that operationalizes exclusion through data, dresses it in the language of verification, and then insulates itself from challenge behind the fiction of algorithmic impartiality.
The constitutional red line thus, becomes absolute. When algorithmic bias perpetuates
discrimination and disentitles citizens from exercising legitimate statutory rights, the tripartite test of the Puttaswamy judgement39 becomes operational. The judgement underscored that any infringement on the right of privacy and dignity under Article 21 must be legal, backed by necessity, and proportional.40 In the same judgment, it was also held that individual dignity is predicated on freedom of choice and autonomy, both of which are precursors to the right to vote. Thus, an AI model that flags over 90 lakh voters as presumptively illegitimate, without
individualized scrutiny, without a defined criterion for exclusion, and without any procedural
mechanism capable of processing that volume before election day, fails all three prongs
simultaneously. It lacks legality because no statute authorizes algorithmic suspension of electoral
participation, it lacks necessity because the scale of flagging was manifestly disproportionate to
any genuine concern about roll accuracy, and it catastrophically fails proportionality because the
least intrusive means—targeted, manual, BLO-led verification—was rendered physically
impossible by the sheer volume the AI generated. If the state delegates its constitutional duties to
opaque AI systems, it undermines procedural fairness and the right to be heard. The voter who
receives no notice, no reasoning, and no meaningful opportunity to challenge their “under
adjudication” status is denied not merely a vote but the fundamental procedural dignity that natural
justice demands.
IV. The German Precedent: Public Examinability as a Democratic Threshold
The issues posed by SIR in Bengal are not merely technical in nature but constitutional, and to address the same, a parallel international precedent could be drawn from the 2009 German elections. The cause of the dispute was the use of computer-controlled voting machines in Germany’s 2005 Bundestag elections. The German Federal Constitutional Court addressed a foundational question that cuts to the heart of this analysis: how much opacity can a democratic process endure before it becomes constitutionally indefensible? The controversy did not arise from established machine malfunction or fraud but had rather cascaded from a concern that the ordinary citizens had no meaningful way to verify whether the process had worked correctly or otherwise. The court simply accepted the contention and held that the use of electronic voting machines, which electronically record and ascertain the voter’s vote, only meets the constitutional Constitutionality of Aadhaar Act: Judgment Summary, SUPREME COURT OBSERVER, requirement of the essential steps of voting, and examination of the result can be done without any specialist reliably.The reasoning of the court was crystal clear. It was the right of the voters to be able to understand, without distinct or detailed knowledge of computer technology, whether their votes cast were recorded in an unadulterated manner as the basis of vote counting or, at any rate, as the basis of a later recount. The legislature in no way is barred from using electronic voting machines in elections if the constitutionally prescribed safeguard of a reliable examination of correctness is duly followed, establishing the principle that all essential steps in the elections are subject to public examinability.
The test is if the democratic process falls below the threshold of public
examinability, it cannot be sustained. The court employed an example to illustrate the same: In conventional elections with ballot papers, there is a high risk of detention, manipulation, or acts of electoral fraud because of human scrutiny of the paper trail, thus serving a prophylactic function. In case of the employment of machines to record and count the votes, there is a wide-reaching effect of programming errors in the software or deliberate electoral fraud committed by manipulating the software, which can only be examined by a specialist and no one to hold accountable. Unlike the known vulnerabilities of paper systems, AI operates through known unknowns, biases embedded in training data and design that cannot be severed from its outputs and that no ordinary citizen can detect or challenge. This is precisely what the German court found constitutionally fatal about EVMs.
However, in the case of AI voter rolls in Bengal, the situation was similar yet way worse. An EVM at least operates at the moment of voting, which the voter can witness, but an AI voter roll makes its determination invisible to the eyes and understanding of the people. The mismatch was foreseeable, but the magnitude of it, with 91 lakh people being deleted from the electoral rolls, was not. The German court struck down EVMs because they failed to bypass the threshold of public examination, and by the same standard, India’s AI voter roll system is constitutionally untenable. If the principle of public examinability applies to how votes are counted, it applies with even greater force to who is permitted to vote at all.
V. Conclusion: Reclaiming Public Examinability in the Democratic Project
The Bengal SIR of 2026 is not just an isolated administrative failure but also a constitutional test that exposes the limits of delegating sovereign functions to opaque algorithmic systems. A tightrope must be drawn between automating convenience and automating the democratic existence of citizens. The state is allowed to deploy technology for day-to-day work to judiciously manage the use of resources, for these are functions where errors are correctable, and no single constitutional moment is foreclosed by a system’s failure. Voter registries and citizenship adjudication belong to an entirely different constitutional category, with sovereignty as a precondition itself.
The constitutional red line, therefore, is absolute when democratic participation is made contingent upon technology that systematically produces discriminatory exclusions against identifiable populations; the tool itself becomes constitutionally impermissible under Article 14. The state cannot invoke an idea of administrative efficiency as a constitutional defense for what is in substance and in effect, the automated disenfranchisement of its own citizens.
Efficiency is not a constitutional value; sovereignty is. Above all else, a democratic process must be publicly examineable and measured against the constitutional values enshrined in Articles 14, 21, and 326 of our Constitution; such a process does not merely fall short. It tramples upon the spirit of equity and democracy and stands as a scathing indictment of everything free and fair elections are meant to represent.
- “AI Software Used to Erase Voters’ Names”— Inside West Bengal’s SIR Chaos, Countercurrents (Apr. 30, 2026), https://countercurrents.org/2026/04/ai-software-used-to-erase-voters-names-inside-west-bengals-sir-chaos/. ↩︎
- Shiv Sahay Singh, Of 90 Lakh Voters Excluded by SIR in West Bengal, 63% Are Hindus, 34% Are Muslims, The Hindu, Apr. 9, 2026, https://www.thehindu.com/elections/west-bengal-assembly/of-90-lakh-voters-excluded-by-sir-in-west-bengal-63-are-hindus-34-are muslims/article70843569.ece. ↩︎
- id. ↩︎
- Gerrymandering | Definition, Litigation, & Facts | Britannica, (May 20, 2026),
https://www.britannica.com/topic/gerrymandering ↩︎ - Supreme Court Issues Directions for Voters Flagged in West Bengal Special Intensive Revision (SIR), SCC Online Blog (Jan. 22, 2026), https://www.scconline.com/blog/post/2026/01/22/sc-directions-west-bengal-sir-electoral-rolls/ ↩︎
- SIR 2.0: Draft Electoral Rolls Out for Bengal, Rajasthan, Goa, Puducherry and Lakshadweep, Deccan Herald (Dec.
2025), https://www.deccanherald.com/amp/story/india/sir-20-draft-electoral-rolls-out-for-bengal-rajasthan-goa-puducherry-and-lakshadweep-3833045. ↩︎