MHT CET 2026: Maharashtra Frames India's First AI-Specific Exam Malpractice Rules

When the Cheating Tool Has an API

For decades, exam malpractice rules in India operated within a fixed vocabulary: chits, mobile phones, impersonators, solver gangs. The threat was human — someone smuggling paper notes into an examination hall, or a more capable person sitting in place of the registered candidate.

That vocabulary is now obsolete.

In June 2026, Maharashtra's State Common Entrance Test Cell issued revised malpractice guidelines for the MHT CET 2026 entrance examinations that, for the first time at a major state level, explicitly named artificial intelligence tools as a prohibited category of cheating method. The guidelines warn that the use of AI tools, hidden cameras, screen mirroring applications, remote-access software, or any other digital cheating technique during computer-based tests may result in immediate cancellation of candidature and legal action under applicable laws.

Maharashtra administers MHT CET across engineering, pharmacy, and agriculture streams, with hundreds of thousands of candidates appearing annually at accredited examination centres. The CET is computer-based, which means every candidate sits at a machine — and every machine is potentially a vector for digitally-assisted fraud.

What AI-Based Cheating Actually Looks Like

The CET Cell's guidelines reflect an emerging threat model that examination administrators across India are only beginning to grapple with. AI-assisted cheating in a computer-based test environment takes several forms:

Screen mirroring to an external solver. A candidate connects their examination terminal's screen output to a remote helper — via a screen-sharing or remote desktop application — who can see the questions in real time and relay answers through an earpiece or a concealed second device. The latency improvements in consumer networking and the commercial availability of remote desktop tools make this operationally feasible in a way it was not five years ago.

AI-powered optical character recognition from a covert camera. A candidate with a tiny lens embedded in clothing or accessories can capture the screen and transmit images to an AI system that reads and answers the questions. Several commercial large language models now achieve competitive accuracy on multiple-choice science and mathematics questions of the type found in engineering entrance exams.

Deepfake biometric bypass. As examination centres deploy facial recognition for identity verification, a new attack surface opens: deepfake video streams that substitute a registered candidate's face for an impersonator's during the biometric check. While not yet widely documented in Indian exam fraud cases, the capability exists and is becoming more accessible.

Remote-access Trojans on examination terminals. If an examination centre's machines are insufficiently hardened — patched but connected to the internet for result submission — it is theoretically possible for malware installed before the examination to allow a remote operator to view or interact with the session.

The CET Cell's guidelines address the first two categories directly. The others represent the next horizon.

What the New Rules Require

The revised malpractice framework for MHT CET 2026 establishes several new obligations for both candidates and examination centres.

For candidates, the rules prohibit:

For examination centres, the CET Cell mandates:

When malpractice is detected or suspected, invigilators must confiscate the device or material, file a written incident report, and require the candidate to provide a written explanation. The case then goes to a CET Malpractice Committee for final adjudication before any sanction is applied — a procedural safeguard that provides due process while taking the decision out of the hands of the individual invigilator.

Why Computer-Based Tests Create a New Security Surface

The shift from pen-and-paper to computer-based testing was motivated by genuine advantages: faster result processing, randomised question sets per candidate, elimination of physical paper logistics, and the ability to hold examinations across multiple days with equivalent paper difficulty using adaptive or parallel forms.

But the computer-based format also introduces risks that paper examinations never had:

Every terminal is an internet-capable device. Even when examination software locks down a machine into kiosk mode, the underlying hardware is a networked computer. Kiosk modes vary enormously in robustness. A poorly implemented lockdown — or a machine with a USB port that hasn't been physically blocked — creates an entry point that did not exist when the examination was conducted with a pen and an answer booklet.

The examination network itself is a target. When tens of thousands of candidates are sitting simultaneously at machines that must retrieve question papers and submit answers, the network infrastructure handling those transactions is under load and under scrutiny. A man-in-the-middle position on that network can, in principle, serve modified content or intercept responses.

AI makes answer-relay faster and more scalable. A traditional solver gang could handle a limited number of candidates — communication was slow and each transaction required a human expert. An AI system that can read a captured question image and generate a high-probability answer can, in theory, serve hundreds of candidates simultaneously with sub-second latency.

None of this means that computer-based tests are more fraud-prone than paper examinations in practice — the removal of physical answer books, the randomisation of question sequences, and the elimination of transit vulnerabilities all represent genuine security improvements. But the threat surface is different, and malpractice rules written in 2015 or 2018 do not map onto it accurately.

What Adequate Examination Centre Security Requires in 2026

Maharashtra's guidelines represent a policy acknowledgement of the new threat model. The operational implementation is harder. Several requirements that follow logically from the guidelines are not yet standard across Indian examination centres:

Network isolation during examination sessions. The safest configuration for a computer-based examination terminal is one that connects only to the examination server on a closed, monitored network, with no internet access for the duration of the session. This requires infrastructure investment at the centre level that most current accreditation standards do not mandate.

USB port disabling and physical sealing. Peripheral ports on examination machines should be disabled at the firmware level and physically covered before sessions begin. This is a straightforward configuration step that prevents a candidate from inserting a device that enables screen capture or remote access.

AI-assisted CCTV review in real time. The CET Cell's requirement for CCTV coverage is necessary but not sufficient. Cameras need to be watched — in real time, with AI-based alerting for specific behaviours: a candidate looking away from their screen repeatedly, head movements consistent with listening to an earpiece, or holding a hand near clothing in a pattern that suggests concealed hardware. Bihar has deployed this capability for government recruitment examinations; it needs to become standard for entrance tests.

Randomised question sequencing per terminal. When every candidate in an examination hall sees questions in a different order, a captured image or screen share provides limited utility — the helper outside does not know which question is being displayed without additional context. Most modern examination platforms support this feature. Centres should verify it is enabled.

Separation of biometric verification from examination session login. Where facial recognition is used for identity verification, the capture should happen at the entry gate — before the candidate is seated at a terminal — not during the session, where a deepfake video stream substitution is more feasible.

The Policy Signal Maharashtra's Guidelines Send

State-level examination bodies rarely lead on exam security policy. The national bodies — NTA for NEET and JEE, CBSE for board examinations — have historically set the tone, with state bodies following years later.

Maharashtra's CET Cell has, in this instance, moved ahead. The explicit naming of AI tools in a malpractice framework — with candidature cancellation as the stated penalty — establishes a legal and administrative basis for action that most state examination rules currently lack. A candidate caught using screen mirroring software during an examination in states that have not updated their rules occupies an ambiguous legal position; Maharashtra has removed that ambiguity.

Other state examination bodies administering computer-based tests — including Rajasthan, Uttar Pradesh, and Tamil Nadu, each of which runs large entrance and recruitment examinations — will need to follow. The technology has moved; the regulatory frameworks need to catch up.

Related Reading