What if the question paper for an exam you are preparing for did not exist until a few hours before the test? Instead of a physical paper prepared by a group of experts, transported covertly between centres and vulnerable to theft or leaks, what if you attempted a paper assembled only shortly before the test?
Infosys co-founder Nandan Nilekani, who heads the Centre’s task force on examination reforms, has proposed using AI to generate question papers shortly before an examination to curb paper leaks. He argued that removing the need for a pre-set paper could eliminate the economic incentive behind leaks. He argued that removing the need for a pre-set paper could eliminate the economic incentive behind leaks. He also called for expanding computer-based testing (CBT) infrastructure to plug leaks in distribution.
“Most Indian exam leaks happen during distribution, days before the test. Lastminute AI assembly only stops distribution leaks if we replace physical shipping with encrypted digital transmission on the morning of the exam,” says Prof Praveen Garimella, Associate Professor of Practice and co-leader at the Centre for Educational Technology and Learning Sciences (CETLS), IIIT Hyderabad.
The Parliamentary Standing Committee on Education, Women, Children, Youth & Sports also recommended a large, continuously updated question bank, with final papers assembled close to the examination using digital tools and vetted by experts.
“AI can help subject-matter experts develop questions aligned with the prescribed syllabus, learning outcomes and difficulty levels. Examination bodies can assemble multiple question papers, reducing dependence on a single paper,” says Doreswamy P, CEO of e-learning technology service provider Excelsoft Technologies.
Experts in the loop
However, experts argue that AI alone cannot guarantee exam integrity. Prof Garimella cautions against asking algorithms to “invent brand-new questions from scratch” on exam day, as this risks factual errors and ambiguous options. Instead, he suggests using AI to help experts build a pre-verified question bank from which a paper can be assembled shortly before the test. Prof Garimella notes that this approach makes human review practical, even at scale.
“Testing 2.4 million students does not mean generating millions of unique questions. Once we separate the scale of testtakers from the scale of the question bank, having a national pool of professors manually review every single question over several months becomes nonnegotiable,” he explains.
Doreswamy adds that reviewing questions is only part of the exercise. Authorities must also ensure that different paper sets test candidates against comparable standards. “The comparability of different question sets must be established using appropriate assessment and psychometric methods,” he says.
An exam system for all
Moving to AI-driven testing with nationwide CBT centres also raises questions of access. Prof Garimella says those unfamiliar with digital interfaces could be disadvantaged, while well-funded coaching centres could use AI tools to anticipate question patterns, giving their students an unfair edge. To address these concerns, he suggests printing decrypted papers at test centres shortly before the exam or deploying lockeddown tablets on offline local networks, alongside free blocklevel practice kiosks. “The AI must be strictly grounded in NCERT textbooks and audited every year so that commercial coaching centres cannot game its patterns,” he adds.
Doreswamy stresses that authorities must account for differences in connectivity, device quality, power availability and technical support across centres. “Distributed LAN-based examination delivery can address connectivity challenges by using a secure local server, reducing dependence on continuous internet access,” he suggests. He adds that examination authorities need visibility into the entire system through regular audits. “A secure examination system should track who accessed or modified questions, when papers were approved and distributed, and how candidate sessions and responses were handled,” he says.
AI-assisted question-paper generation offers a new approach to tackling exam leaks in India, but it cannot work as a standalone solution. Its effectiveness will depend on expert oversight and the security of the systems used to deliver examinations.