As AI-linked job postings are projected to grow steeply, recruiters are expecting AI fluency as a baseline skill for entry level jobs. In order to prepare graduating students to seize opportunities, Higher Educational Institutions (HEIs) are racing to impart hands-on AI skills to students by integrating AI seamlessly not only into the curriculum but also in the pedagogy. This brings us face-to-face with an important question: should students be allowed to use AI in classrooms and examinations? What are the concerns of stakeholders? How to address them?
(Sign up for THEdge, The Hindu’s weekly education newsletter.)
The stakeholder tug-of-war: fear, confusion and free-riding
For students, AI platforms are 24/7 personalised smart assistants, ready to answer any question at any time. A recent 2026 survey report reveals that AI has transitioned from a novel technology tool into a daily baseline utility for students and that an overwhelming 82% of Indian college students actively utilise generative AI for their coursework, homework and test preparation, significantly exceeding the global average of 62%. But this convenience comes with a side of operational confusion. Many admit their struggle with verifying whether an AI’s answer is actually correct. More importantly, a significant amount of mental energy is spent on digital camouflaging: how to mask AI assistance to look like original work.
Upcoming webinar: Is India a literate country?
Teachers, understandably, are losing sleep over this. Despite sophisticated institutional guardrails, AI has arguably become the easiest mechanism for academic malpractice ever invented. Malpractice now ranges from earpieces and secondary devices used during online exams to a subtler brand of academic free-riding. Here, AI generates plausible-looking citations, fake experimental data or an entire group project, while the student claims equal credit. The traditional take-home assignment is dead.
When a flawless, syntactically perfect essay can be conjured in a few seconds, the traditional grading rubric crumbles. The pattern across all these cases is clear: detection alone is a losing, ever-escalating arms race. Every new AI detector eventually meets an even sharper evasion tool. Just as we didn’t stop Maths testing because of calculators, we cannot stop evaluation of essay writing because of AI. We must change how we test. The educators harbour a much deeper fear: the catastrophic erosion of a student’s underlying critical thinking abilities.
For academic administrators, the dilemma is how to prepare students for AI-driven jobs, at the same time addressing issues like code of conduct, infrastructural scaling and cyber security risk.
Finally, policymakers recognise that completely banning AI in the HEIs will produce graduates that may be unemployable in the new corporate world. While they are keen to encourage rapid technological innovation to remain competitive globally, they also want maintain strict guardrails to protect public data privacy, ensure algorithmic safety and maintain national educational integrity.
Across all the stakeholder groups, the most consistently cited obstacle is not a principled opposition to AI but a vacuum of clearly communicated policy and guidelines addressing the concerns of the diverse stakeholders.
Case studies on AI integration and lessons learnt
The global universities that embraced AI early-on have generally tried to integrate it into the syllabus rather than the exam hall. Arizona State University, Wharton and Oxford have all deployed campus-wide ChatGPT Edu access and used it for tasks such as guided reflection projects and language-practice tutors. The University of Michigan rolled out a secure, custom, privacy-compliant AI platform for all students and faculty.
The IIMs have pioneered AI-co-piloted case studies. Students are explicitly encouraged to query LLMs to gather baseline competitive research, conduct market analysis and generate initial financial projections. However, the final grade is not based only on the final answer; instead, students are evaluated on their ability to critique the AI’s outputs, spot structural hallucinations and defend their personal strategic conclusions in Viva Voce before a faculty panel.
IIT Madras has selectively integrated AI tools into foundational coding courses. Students use AI to debug complex programs in real time. The focus shifts entirely away from memorising syntax to mastering high-level system logic and computational problem-solving. Institutions like IIIT Delhi and IIM Nagpur allow LLMs in examinations and assignments in select departments, on the condition that students submit the prompts they used alongside their answers.
The experimental implementations of AI at various institutions have provided three clear, foundational lessons. Mandating absolute bans on AI tools is futile. Students can easily bypass campus firewalls via private data networks and mobile devices. Bans simply drive the technology underground, creating an environment of unequal access where tech-savvy students continue to use AI secretly while honest students are left at a disadvantage.
Assessments must shift away from evaluating the final answer to evaluating the cognitive journey and must focus on the students’ prompt engineering strategies, their critical analysis of AI outputs and their ability to defend their approach during the interactive viva voce. Relying on publicly available free commercial versions of AI tools exposes intellectual property and sensitive student data to commercial exploitation. Institutions must invest in secure, closed campus-wide AI instances that guarantee absolute data confidentiality.
Teaching students to use AI effectively
If active, constructive engagement is the goal, prompt engineering is the discipline that gets students there. It involves externalised thinking, stating the goal, context, constraints and the desired output clearly enough so that the student has clarity on the problem to be solved. Students can start learning a small set of named, reusable prompt structures—specifying context, asking for stepwise reasoning, requesting a particular output format, assigning the model a persona or role—and are then asked to submit their prompts alongside their finished work, so a teacher can evaluate the thinking process, not just the polished output. A vaguely specified prompt is one of the most common causes of an AI hallucination, so teaching students to write a precise, well-constrained prompt is itself a form of error prevention, not merely a productivity trick.
Will AI prevent cultivation of higher order thinking skills?
Bloom’s taxonomy pyramid on learning moves up from remembering and understanding, through applying and analysing, up to evaluating and creating. Generative AI is disconcertingly good at the lower rungs, recalling facts, summarising, producing a first draft, which may be the reason for the argument that classroom time and assessment weight should shift almost entirely to the top three levels. Used well, AI can be a sparring partner for higher-order work: a student can ask it to defend a counter-argument or stress-test a business plan, and then has to evaluate, correct and improve on what comes back. IIM Sambalpur’s pilot is an instructive example: it uses an AI platform to score live MBA case-discussion participation against Bloom’s six cognitive levels, returning detailed feedback on where each student’s contribution actually sat on the ladder.
Empowering educators: Need for a teacher-training framework
We cannot expect all students to use AI responsibly if our teachers do not understand how the technology works. Training our educators is the single most critical step in this transformation. Institutional training frameworks must target three fundamental domains:
First, teachers must undergo rigorous technical and prompt literacy workshops. They need to understand what happens beneath the hood of an LLM and why models hallucinate data with high confidence. Educators must learn to write sophisticated prompt frameworks, use advanced retrieval techniques and effectively leverage AI as a productivity tool rather than viewing it as a mysterious black box.
Second, teachers must master the art of AI-resilient assessment design. Training programs must teach faculty how to craft assignments that machines cannot easily solve. This includes focusing on hyper-local contexts, contemporary case studies occurring after a model’s data cutoff, or designing multi-stage workflows where students submit initial handwritten concept maps, detailed logs of their AI prompts and a final critical synthesis.
Finally, training must emphasise AI ethics and bias mitigation. Educators must be trained to identify systemic gender, racial and cultural biases embedded within western-trained LLM models. They must pass this critical scepticism down to their students, ensuring that the classroom uses AI as an analytical trustworthy partner rather than an unguided oracle.
Global paradigms: how developed nations are governing AI in classrooms
While India stands at the crossroads, developed nations have already moved from reactive anxiety to proactive systemic governance, building robust models that ensure ethical, responsible and effective AI deployment.
The European Union (EU) AI Act classifies AI systems used in educational and vocational institutions as ‘high-risk’. This legally mandates strict human oversight, algorithmic transparency and rigorous data protection. Furthermore, its updated regulations place mandatory AI literacy obligations on the institutions, which are legally required to ensure both educators and students understand how these models function, including their inherent biases.
Singapore pioneered an ‘AI-in-Education (AIEd) Framework’ through its Ministry of Education, anchoring technology into its Masterplan 2030, embedding custom, guided AI tools directly into its student portal.
Australia deploys specialised, curriculum-aligned AI sandboxes designed explicitly to prompt critical thinking and error-validation.
A framework used internationally to enable effective governance is the AI Assessment Scale, developed by Perkins, Roe and Furze. It sets out five graded levels of permitted AI use, from ‘No AI’ through to full ‘AI Exploration’, that a teacher chooses deliberately for each specific task, intended as the start of an explicit conversation with students about whether AI use is appropriate for that piece of work, rather than catching or detecting violations after the use.
The regulatory roadmap for India
To ensure India’s massive student population converts this technological disruption into a global advantage, regulatory bodies like the University Grants Commission (UGC) and the All India Council for Technical Education (AICTE) must move away from reactive warnings to establish a proactive, systemic regulatory framework relevant for India.
The government should fund the development of dedicated, low-cost, sovereign academic LLMs trained on Indian curricula and multilingual datasets. These secure models should be provided free to all public universities, protecting student data privacy while ensuring absolute equity of access.
The path forward: A balanced synthesis
Ultimately, the question confronting Indian education is not whether we should allow AI in our classrooms and exam halls. The real question is how to manage this transition with proactive vision or reactive fear. By shifting the focus from enforcing futile bans to building advanced critical thinking, re-engineering assessment methods and providing deep training for our educators, India can transform this artificial intelligence disruption into a powerful catalyst for leveraging human intelligence. The classroom of the future should not be a conflict between students and AI technology; it must be a collaborative space where the advanced AI technology elevates human potential to new heights.
(Prof O.R.S. Rao is the Chancellor of the ICFAI University, Sikkim. Views are persona.)
Remove