Can AI Stop India's Exam Paper Leaks? Here's What Works

Key takeaways

• Exam paper leaks happen because a finished question paper exists days or weeks before the exam, and that paper has a price on the black market.
• Nandan Nilekani's proposal removes the paper itself: keep a huge, secure question bank and assemble each paper only minutes before the exam on a computer.
• AI should assemble the paper, not write the questions. Teachers write and grade the questions; AI picks a balanced set in real time.
• It will work only if the question bank is locked down, refreshed every year, fair on difficulty and tested on smaller exams before NEET.

What if there were no question paper to leak?

That was the question on CNN-News18's The Breakfast Club on 14 October 2026, and I joined the programme to talk about it. India's paper leak crisis has put the future of millions of students at risk, from the NEET UG controversy to UGC-NET cancellations and CBSE paper discrepancies. Every leak means a re-test, months of uncertainty and a little less trust in the system.

You can watch the full discussion on CNN-News18. Here is what I said, and why I think it matters.

What Nandan Nilekani proposed to stop paper leaks

Stop printing a fixed question paper weeks in advance, and assemble it from a large, secure question bank just before the exam.

Nandan Nilekani, Infosys co-founder and chair of the high-powered task force on examination reforms, put it simply on the programme: "the moment you have a question paper, then that creates an economic incentive" for it to be sold. His answer is to build a massive bank of questions, generate the paper for that day only at the last minute, and move exams to computer-based testing.

He also pointed to the network India already has, about 1.5 lakh locations where a young person can sit a computer-based test, and said it needs to grow.

My take: He isn't trying to kill the exam. He's trying to kill the black market. A paper that doesn't exist until the morning of the exam can't be sold the night before.

We solved the same problem at a smaller scale

When our own test questions leaked, we stopped having "a paper" and started generating each test in real time from a very large pool.

At Techila, our interview and assessment questions once ended up on the internet. Candidates walked in already knowing the answers. About a year ago we built a pool of hundreds of thousands of questions, and each test is now generated on the computer in real time. That became Talent IQ.

The logic is simple. Someone can memorise 20 or 30 leaked questions. Nobody can memorise hundreds of thousands. We also made the test adaptive: depending on how a candidate is doing, the next question gets harder or easier.

But I was clear on air: doing this for NEET, with more than 22 lakh candidates sitting the same exam in many languages, is a different order of difficulty.

AI should assemble the paper, not write it

Today's AI still hallucinates, so teachers should write and grade the questions, and AI should only pick and package them.

Think of it like a kitchen. The cooks are still human: teachers write tens of thousands of questions over time and check every one. AI is the packer. On exam day it picks a balanced set from that bank, in real time, for each centre or each candidate.

That split matters. AI can help check difficulty, spot duplicates and build equivalent sets. But letting AI write the questions for a national exam today would trade a leak problem for an accuracy problem.

The hard parts: security, fairness and languages

Removing the printing press moves the risk from a printing room to a database, so the hard work shifts to security, fairness and language quality.

A digital vault with footprints

If the paper lives in a system, the system becomes the target. The question bank needs to sit in a digital vault where every access leaves a footprint, overseen by trusted institutions and senior security experts, so that anyone who opens it can be traced.

Refresh the bank every year

Coaching centres are smart. They collect questions from their students after every exam. If the same pool circulates for a decade, it leaks slowly instead of all at once. The bank has to be refreshed every year.

Prove that every paper is equally hard

The first complaint with any per-student paper is "my set was harder than theirs". The fix is a complexity score for every question, say from 1 to 5, set by teachers and checked by AI. Each paper is then built to the same mix, for example 30% level 1, 30% level 2 and so on. That makes fairness something you can show, not just claim.

Regional languages are the weak spot

NEET runs in many regional languages, and AI is not yet good enough across all of them. The same question can come back in different forms, or a translation can change its meaning. This needs careful testing before anyone relies on it.

My take: The technology for last-minute assembly exists today. The difficult parts are the vault, the fairness scoring and the regional languages, and those take time to get right.

Start small: pilot it before NEET

Run the system on smaller exams for 6 to 12 months, fix what breaks, then move to NEET.

My view on the programme was that NEET is too big a system to be the first test. A failure there would hurt more than 22 lakh students at once. Start with smaller entrance exams over the next 6 to 12 months, measure fairness and security, expand the testing centres, and then scale.

The idea is right. How it's rolled out will decide whether it works.

What should happen next:

• Build the question bank now, with teachers writing and grading every question.
• Put the bank in an audited digital vault where every access is logged.
• Agree on a published difficulty scale, so every candidate's paper can be shown to be fair.
• Pilot last-minute assembly on smaller exams for 6 to 12 months, including regional languages.
• Grow the computer-based testing network so rural students get the same access as city students.

Frequently asked questions

Can AI stop exam paper leaks in India?

AI can help a lot, but not alone. The leak disappears when there is no fixed paper to steal: a large human-written question bank, with the paper assembled by software minutes before the exam. AI's job is to pick a balanced set in real time, not to write the questions.

What is Nandan Nilekani's proposal for exam reform?

He proposed replacing fixed question papers with a massive, secure question bank, generating each day's paper only at the last minute, and expanding computer-based testing across India's roughly 1.5 lakh test locations. The aim is to remove the economic incentive to leak a paper.

Is last-minute question paper generation fair to every student?

It can be, if every question carries a difficulty score set by teachers and checked by AI, and every paper is built to the same mix of difficulty levels. That makes fairness measurable instead of a matter of opinion.

Should NEET switch to this system straight away?

No. NEET is too large to be the first test. A safer path is to pilot the system on smaller entrance exams for 6 to 12 months, fix the problems, and then roll it out to NEET.