Author: Renju Abraham, a LL.M. student at the Central University of Punjab.
Abstract
India’s National Eligibility-cum-Entrance Test for Undergraduate medical admission (NEET-UG) has, in 2024 and again in 2026, been struck by paper leaks that forced the National Testing Agency (NTA) to cancel and re-conduct examinations, affecting millions of aspirants. In response, the NTA deployed an unprecedented artificial-intelligence surveillance apparatus for the 2026 re-examination, combining live-feed AI monitoring, mass biometric authentication, and tens of thousands of electronic jammers. This article asks whether that response is constitutionally sound. Its central thesis is that AI-based examination surveillance is not a neutral technical fix for institutional failure but a constitutional event, implicating equality under Article 14, the right to livelihood under Article 19(1)(g), and privacy, dignity, and due process under Article 21. Adopting a doctrinal and comparative method, the article situates NTA’s response within Supreme Court jurisprudence on privacy, proportionality, and natural justice, contrasting it with interventions in the United States, the United Kingdom, and the European Union. It finds the NTA’s surveillance architecture to be a demand-side response to what was, in both crises, a supply-side failure of custody, rendering it disproportionate under the necessity limb of the Puttaswamy test. The article concludes with ten constitutionally grounded recommendations, including mandatory algorithmic audits, a statutory right to explanation, and judicial review of AI-assisted decisions, aimed at reconciling examination integrity with constitutional fairness.
Introduction
On May 3, 2026, nearly 2.27 million young Indians sat for the National Eligibility-cum-Entrance Test for Undergraduate medical admission, the single most consequential examination gateway in the country. Within days, the National Testing Agency (NTA) cancelled the exam after investigators found troubling overlaps between a circulating “guess paper” and the actual question paper. A re-examination on June 21, 2026 came with a security apparatus unprecedented in Indian examination history: tens of thousands of electronic jammers, over a lakh CCTV cameras across roughly 95,000 rooms, biometric face-authentication for every candidate, and, for the first time at this scale, artificial intelligence analysing live camera feeds to detect malpractice in real time.
This is not merely a story about examination security, but about a constitutional democracy, embarrassed by repeated institutional failure, reaching for the most powerful surveillance technology available and turning it inward upon its own citizens at the most high-stakes moment of their young lives. The central argument here is simple: AI-based examination surveillance is not a neutral technical fix for a governance failure but a constitutional event implicating equality, privacy, dignity, and due process, and unless anchored in enforceable transparency and accountability standards, it risks curing a legitimacy crisis with a legality crisis.
Background
NEET-UG’s troubled recent history did not begin in 2026. In 2024, allegations of a systemic paper leak originating in Hazaribagh and Patna reached the Supreme Court. The Court, led by then Chief Justice D.Y. Chandrachud with Justices J.B. Pardiwala and Manoj Misra, declined to order a re-test for the roughly 24 lakh candidates who had appeared, holding the leak was not systemic enough to have contaminated the entire exercise, since the CBI had traced beneficiaries to a limited number of candidates. The Court nonetheless expanded the mandate of a Centre-appointed committee, headed by former ISRO Chairman K. Radhakrishnan, to recommend structural reforms.
Two years later, history repeated itself with uncomfortable precision. The 2026 NEET-UG examination was cancelled on May 12, ten days after it was held, when investigations revealed overlaps between a pre-circulated guess paper and the actual paper administered to over 2.27 million aspirants. When the matter reached the Supreme Court, a bench led by Justice P.S. Narasimha remarked pointedly that the authorities had not learnt the lessons of 2024, and issued notice on petitions including one by the United Doctors’ Front seeking the NTA’s dissolution and a re-examination supervised by a retired Supreme Court judge.
In response, the NTA told the Court that cancelling the exam itself demonstrated its commitment to integrity, and detailed corrective measures including dedicated printing supervision and a documented chain-of-custody for papers transported under central armed police escort. It is against this backdrop of institutional distrust that AI surveillance entered the picture not as a considered policy choice, but as an emergency response to scandal, deployed at a scale that outpaced the legal framework meant to govern it.
The Emerging Legal Issues
The shift to algorithmic monitoring raises four distinct legal questions Indian law has yet to answer.
First, what exactly is the AI doing, and under what authority? NTA officials describe the system as capable of detecting candidates communicating with one another and identifying attempts to carry mobile phones during the examination. But “the AI flagged it” is not a legal standard; it is a claim of technical capability. No published regulation defines its decision thresholds, false-positive rate, or the human-review process preceding a disqualification.
Second, what happens to the biometric data collected? Reported measures for 2026 included face-authentication for every candidate, with biometric personnel roughly doubled for the retest. Facial and behavioural data are sensitive personal data; their retention and secondary uses remain largely undisclosed. Notably, many NEET aspirants are minors at registration, engaging the heightened protections Parliament reserves for children’s data.
Third, what recourse exists for a wrongly flagged candidate? A false positive here can end a medical career before it begins, yet no NTA circular grants a right to see the basis for a flag, or an appeal resembling a natural-justice hearing.
Fourth, is centralised AI surveillance proportionate to the actual mischief? Both scandals originated as paper leaks in printing and transport, not candidates cheating inside the hall. Training AI cameras on every candidate addresses a different, lesser risk than the one that actually materialised twice.
Constitutional Analysis
Article 14: equality before law. Any classification or adverse action must satisfy the twin test of intelligible differentia and rational nexus. An opaque AI system flagging candidates on undisclosed parameters risks arbitrary, unequal treatment; identical behaviour may be assessed differently owing to camera angle, lighting, a regional accent misread as suspicious speech, or a disability misread as anomalous movement. Since E.P. Royappa v. State of Tamil Nadu, arbitrariness has itself been treated as antithetical to Article 14; an unreviewable algorithmic judgment risks institutionalising exactly that arbitrariness.
Article 19(1)(g) the right to pursue an occupation. For a NEET aspirant, the examination is the sole gateway to a medical career. A wrongful AI-triggered disqualification directly restricts this right, and under Article 19(6) any restriction must be reasonable, requiring the mechanism imposing it to be explicable and challengeable, not a black box.
Article 21: life, personal liberty, and its expanding penumbra. This is where the deepest tension lies. Privacy was declared fundamental, intrinsic to liberty and dignity under Article 21, by a unanimous nine-judge bench in Justice K.S. Puttaswamy (Retd.) v. Union of India. This built on older authority recognising bodily and informational privacy, and on the Court’s holding that compelled physiological testing implicates liberty and the privilege against self-incrimination a caution equally relevant to compulsory biometric capture in an examination hall. The Aadhaar judgment refined this into a proportionality test requiring a legitimate aim, a rational means-aim connection, necessity, and a proper balance between rights impact and benefit. Applied to blanket AI surveillance of every candidate’s face and body for the exam’s duration: the aim is legitimate and the connection plausible, but necessity is doubtful, since the demonstrated failures originated in printing and transport, not inside the hall, and secure logistics with conventional invigilation could plausibly address that failure without maximal surveillance. The balance tilts unfavourably where surveillance is maximal, but the demonstrated threat is narrow.
Dignity and the chilling effect. Continuous monitoring of eighteen-year-olds, many unfamiliar with camera-based scrutiny, can itself become a source of psychological pressure undermining the fairness it claims to protect. Dignity under Article 21 means freedom from being treated as a suspect by default, not merely from physical restraint.
Natural justice, due process, and administrative fairness. Maneka Gandhi v. Union of India transformed “procedure established by law” under Article 21 into one that must be fair, just, and reasonable. Settled administrative-law principles require that any decision capable of civil or professional consequence give the affected person notice and a genuine hearing before an adverse order. A wholly automated flag-to-cancellation pipeline, without meaningful human review, would offend audi alteram partem, sharpened by the Court’s insistence that Article 21 protects the means of earning one’s livelihood with dignity, not mere animal existence.
Critical Analysis
The government’s defence is not without force. India’s examination ecosystem has been repeatedly compromised by organised, sometimes transnational, cheating syndicates, and 2024 and 2026 show human-only oversight has failed twice in two years. The state has a compelling interest in a process allocating medical seats through merit.
Yet the critical assessment offered here is that the NTA has committed a category error: it has answered a supply-side failure leaked papers and a corrupted printing and transport chain with a demand-side surveillance solution that watches the candidates themselves. The measures in the NTA’s own affidavit dedicated printing supervision, banned devices, documented chain-of-custody are the genuinely responsive fixes, and none require AI surveillance of examinees. The AI camera layer functions less as a proportionate remedy than as a reassuring performance of control for a public and judiciary that has lost patience with a legitimate political need, but not, without more, a constitutionally sufficient justification for mass biometric surveillance tied to a citizen’s livelihood. This is not an argument against better invigilation; it is an insistence that surveillance’s scale must track the actual locus of risk.
Evidence and Recent Developments
The scale of the 2026 apparatus is instructive: the June re-examination reportedly spanned roughly 5,440–5,454 centres in 551 Indian cities and 14 abroad, with CCTV across nearly 95,000 rooms monitored through over 1.38 lakh cameras, AI-based real-time analysis, more than 51,000 jammers, close to 48,500 biometric personnel, and roughly 6,700 observers. The Education Minister separately announced Air Force deployment to transport question papers, while the CBI’s investigation led to a growing number of arrests.
Judicially, the Supreme Court has not yet ruled on the constitutionality of AI surveillance itself; its 2026 intervention has focused on institutional accountability, issuing notice on pleas seeking the agency’s dissolution, and directing an affidavit on compliance with the 2024 reforms. This leaves a live gap: the executive has moved swiftly on surveillance technology, while judicial oversight remains trained on institutional design rather than the rights of the individual candidate facing an algorithmic accusation.
Counter-Perspectives
Three counter-arguments deserve engagement. First, AI reduces human bias and corruption, since a centrally reviewed camera feed is harder to suborn than a bribed invigilator, a genuine advantage. Second, consent is arguably implicit: candidates who register know the terms and submit to biometric checks, much as air travellers accept security screening. But this understates the asymmetry: for an aspiring doctor, NEET is the only door, making “consent” more formal than real.
Third, defenders note AI merely flags while humans decide, but without published review protocols this is unverifiable, and comparative experience suggests “human in the loop” safeguards can become rubber stamps under volume pressure, mirroring judicial scepticism elsewhere toward opaque restrictions on freedoms unaccompanied by adequate procedural safeguards.
Comparative International Perspective
India is not alone in confronting this tension. In the United States, a federal court addressed a strikingly similar question in the context of pandemic-era remote proctoring. In Ogletree v. Cleveland State University, a federal judge in Ohio held that a “room scan” required by proctoring software amounted to an unconstitutional search under the Fourth Amendment. The student’s privacy interest in his home outweighed the university’s interest in deterring cheating, given the home’s special protection and the absence of a real alternative, even though the intrusion lasted less than a minute. The remedy was narrow but real: no room scan without a reasonable alternative, resonant, in substance, with the proportionality logic of Puttaswamy.
The United Kingdom offers a closer analogue on facial recognition. In R (Bridges) v. Chief Constable of South Wales Police, the Court of Appeal held that a police force’s live facial-recognition trials were unlawful for lack of an adequate legal framework governing who could be watch-listed, where it could be deployed, and how bias would be audited. The parallel is direct: NTA’s system likewise operates without any published framework for its logic, deployment boundaries, or bias-testing protocol.
The European Union has taken a more systemic approach. Under the EU AI Act’s Annex III, AI systems used to determine admission to, or evaluate outcomes in, educational institutions are listed as “high-risk,” triggering mandatory risk management, documentation, human oversight, and registration the binding, upfront classification India’s notification-driven approach still lacks.
Recommendations
A durable reconciliation of examination integrity with constitutional fairness requires a statutory architecture built for high-stakes public examinations using AI. The following ten measures would constitute such an architecture.
- Mandatory independent AI audits before deployment. No AI system should be deployed at NEET’s scale without certification by an auditor independent of the NTA and the vendor, since the NTA is currently both deployer and assessor of its own technology, a conflict of interest. This is grounded in the Aadhaar test’s necessity limb: actual performance, not vendor claims, must justify the intrusion. It is implementable on the EU AI Act’s conformity-assessment model, with MeitY empanelling auditors and publishing a registry.
- Annual AI transparency reports by the NTA. The NTA should be required to publish, after each cycle, the parameters the AI monitors, flags raised, flags overturned on review, and their demographic spread, correcting the asymmetry under which even the Supreme Court has relied on NTA affidavits for basic facts. This is grounded in the transparency norms read into Article 14, and is implementable through an amendment requiring publication within a fixed period.
- Mandatory Algorithmic Impact Assessments (AIAs). Before introducing any AI system into a public examination, the NTA should publish an AIA identifying foreseeable rights impacts and mitigation measures, akin to an environmental assessment but for algorithmic harm, since such technology is deployed reactively, without anticipatory analysis. This is grounded in the proportionality doctrine’s demand that necessity be shown before deployment, and is implementable as a precondition to procurement, enforced by withholding sanction until filed.
- A statutory Right to Explanation for AI-assisted decisions. Every candidate adversely affected by an AI flag should have an enforceable right to a clear explanation in plain language, since challenging a decision is meaningless without understanding it. This is grounded in the audi alteram partem principle and the fair-procedure reading of Article 21, and is implementable as a dedicated NTA provision with a fixed timeline for reasons.
- Judicial review of AI-assisted administrative decisions. Any disqualification based on an AI flag should be justiciable before the jurisdictional High Court under Article 226, with courts empowered to call for algorithmic logs, since finality must never be conflated with infallibility. This flows from judicial review as part of the basic structure, and is implementable by regulations preserving this remedy and mandating retention of logs.
- Mandatory human oversight before any adverse action. No candidate should face disqualification on an unreviewed machine output alone; a trained officer must independently review the footage and concur first, since the false-positive risk of behavioural AI is unacceptably high when the stake is a medical career. This is grounded in the fair-hearing requirement of natural justice, and is implementable through a dual-signature protocol: AI flag plus human affirmation.
- An independent grievance redressal mechanism. A body external to the NTA a standing tribunal or ombudsperson should hear AI-flag complaints within a fixed period, since a complainant cannot expect impartial review from the agency whose system generated the flag. This reflects the principle that natural justice is undermined when the decision-maker lacks independence, and is implementable by notification, with binding power to annul a wrongful disqualification.
- Periodic bias testing and third-party certification. The AI system should undergo recurring, independent bias audits across gender, region, disability, and language, certified periodically rather than once at launch, since algorithmic bias shifts as demographics change. This engages Article 14’s guarantee against disparate-impact discrimination, and is implementable on the EU model, adapted to an Indian certifying authority.
- Robust data protection, retention, and deletion policies. Biometric data should be retained only as long as needed for grievance resolution and audit, encrypted throughout, then irreversibly deleted, since indefinite retention of lakhs of candidates’ data, many minors, is a security risk disproportionate to any benefit. This is grounded in the child-data obligations under the Digital Personal Data Protection Act, 2023, and is implementable by designating the NTA a data fiduciary.
- A comprehensive statutory framework for AI in high-stakes public examinations. Parliament should enact a dedicated statute rather than leave the field to ad hoc notifications, incorporating the nine measures above as binding standards, jointly enforced by the Ministry and a technical regulator, since the notification-driven approach has proven reactive rather than preventive. This is grounded in Parliament’s power to give legislative content to guarantees courts have enforced only case by case, and is implementable as a standalone AI Governance Act.
Conclusion
The NEET-UG saga of 2024 and 2026 is, at its core, a story of a state institution trying to restore public trust after repeated failure and reaching, understandably but too hastily, for the most powerful monitoring tools available. AI surveillance can be legitimate, but only if its design honours the constitutional architecture built around Articles 14, 19, and 21: proportionate, transparent, and subject to a genuine human check before it extinguishes a young person’s right to a livelihood. The leaks of 2024 and 2026 were failures of custody in printing rooms and transport vans, not of candidate honesty inside examination halls. A democracy that surveils its citizens more, rather than governing itself better, has not restored trust; it has merely relocated the deficit. NEET-UG’s future legitimacy will be measured not by how many cameras watch candidates, but by how transparently the watchers answer for what they see.
