CBT, question vault, and GenAI for NEET: solutions or new risks?

Over 22.79 lakh students appearing for the NEET-UG re-examination across 5,440 centres on June 21, 2026. Photo. Ragu R / The Hindu Following the NEET-2024 fiasco, the Dr. Radhakrishnan Committee recommended 101 reforms, reportedly approved without prior pilot studies or an evidence-based action report.

Despite the National Testing Agency (NTA) being tasked with implementing them, NEET-2026 culminated in another crisis, with the cancellation of the examination and a subsequent re-examination. The formation of a High-Powered Task Force (HPTF) led by former UIDAI chairman Nandan Nilekani marked another shift in approach.

Mr. Nilekani, at the Bloomberg New Economy Forum 2026, gave suggestions to prevent paper leaks by creating a secure question vault and last-minute question paper assembly, with GenAI also being considered as a possible enabler.

He also said the panel is studying the feasibility to conducted all exams using CBT (Computer-Based Test). NEET has historically been conducted in the Pen-and-Paper (PnP) format, although CBT is already used in many examinations conducted by the NTA and other bodies.

Following successive NEET fiascos, the Union government has repeatedly announced a shift to CBT, also advocated by the Dr. Radhakrishnan Committee. CBT may reduce risks associated with printed papers, OMR tampering, and physical transportation, but does not eliminate underlying vulnerabilities.

Both formats remain exposed to insider access, conflicts of interest, and human involvement in question setting, review, and translation. CBT also introduces digital risks related to question banks, software vendors, server access, cybersecurity, and examination-center operations.

CBT raises further concerns about equity and operational reliability. Rural candidates and those with limited computer exposure may be less comfortable or slower at navigating digital examinations than their urban or more privileged counterparts.

Technical glitches and system failures have also been reported at examination centres, including in metropolitan areas, potentially disadvantaging candidates through no fault of their own. Unless institutional, technical, and equity risks are addressed, CBT may merely shift vulnerabilities from physical to digital systems while disadvantaging candidates affected by the digital divide.

Mr. Nilekani’s suggestions challenge the conventional practice of preparing a question paper well in advance. Instead, it envisages maintaining a large, continuously updated pool of vetted questions in a secure digital vault, from which papers would be assembled shortly before the examination and securely transmitted to centres.

The objective is to minimise leaks in advance by reducing the time available for access, copying, or selling the final paper. For the projected 16–20 NEET sessions, each accommodating 1.2–1.5 lakh candidates to cover 24 lakh registered candidates, the system would require a large, diverse question pool, backup sets, and a carefully defined blueprint covering syllabus units, question types, and difficulty levels.

MCQ options would need appropriate ordering. Every question would require expert review, translation where necessary, independent solving and verification, with answer keys validated before use. Papers across sessions would also need comparable difficulty.

A question vault, however, is not inherently leak-proof. Its security would depend on robust access controls, separation of responsibilities, traceable selection and transmission, and independent oversight. A large question pool may reduce the risk of advance access to the final paper, but it also creates a larger repository of sensitive material.

Past NEET controversies have highlighted alleged insider involvement and compromised networks; technological safeguards alone cannot address institutional vulnerabilities or collusion. The method used to assemble each paper is also crucial.

Conventional software could select questions from the vault using predefined criteria and a table-driven process, without requiring GenAI. This could limit unnecessary information exchange and reduce additional security risks associated with AI systems.

GenAI-based question paper generation is a distinct, potentially complementary approach: it can create a question paper but may introduce further concerns about prompt and output security, data exposure, and, optionally, translation accuracy.

The suggestions focus on last-minute assembly from a vetted question pool; GenAI’s precise role remains to be formalised in the task force’s final report. The complete system—from question creation and validation to paper assembly, access management, and delivery—must be independently audited to establish its security, reliability, and fairness.

Conducting NEET through CBT may require multiple sessions, depending on infrastructure and operational design. Each session would require a different question paper, making comparable difficulty a critical challenge, potentially compounded if GenAI is used to generate questions or papers.

Perfectly identical difficulty cannot be guaranteed. Raw marks across sessions may therefore not be directly comparable, necessitating statistical normalization, potentially using percentile scores. However, normalisation cannot eliminate all uncertainty.

Its reliability depends on question-paper comparability, valid statistical assumptions, and a transparent methodology. The stakes are particularly high in NEET, where even small changes in rank can determine medical admission.

Unlike JEE Main, which primarily serves as the qualifying examination for JEE Advanced, NEET scores directly determine eligibility and merit for undergraduate medical admissions. With millions competing for limited seats, even modest adjustments to normalised scores can affect candidates near the admission cutoffs at premier institutions.

Normalisation cannot compensate for poorly calibrated or fundamentally unequal question papers. Before...