India is Scaling AI in Education. It Should Scale the Guardrails Too
The economic survey 2025-26, tabled at the end of January, has an entire chapter dedicated to the implications of Artificial Intelligence (AI) on the Indian Economy. Economic Survey 2025–26, Ministry of Finance, Government of India It delves into the implications of AI in the labour market, Data Governance, and safety. However, within all this, if one reads the chapter thoroughly, there are indications of a unique problem that AI brings with itself. The Economic Survey cites research from the Massachusetts Institute of Technology and Microsoft to lay emphasis on the fact that growing dependence on Generative AI for creative and writing tasks is contributing to cognitive decline and deterioration of critical thinking abilities amongst students.
The MIT Media Lab study, which the survey cited, was led by Nataliya Kosmyna et al. and published in June 2025. “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task” – MIT Media Lab The study used electroencephalography on 54 University Students and found that those who wrote essays using Generative AI tools, showcased 55% reduced brain connectivity, when compared to students who wrote their essays unaided. To make things worse, 83% of the participants who had used AI, were unable to recall a single sentence from their AI generated content. This study led the researchers to coin a term known as “Cognitive Debt,” describing a pattern where short term productivity gains eventually accumulate into long term memory deficits. The other study, led by Hank Lee at Microsoft Research, surveyed 319 knowledge workers and concluded that a high confidence level in AI correlated with limited critical thinking, as workers increasingly shifted from actual problem solving to a mere passive verification of AI outputs. “The Impact of Generative AI on Critical Thinking” – Microsoft Research / CHI 2025
These studies are not isolated and represent a growing literature of research dedicated towards the impact of AI on human cognition. For instance, a large randomised controlled trial published in the Proceedings of the National Academy of Sciences by Hamsa Bastani et al. at the Wharton School tested approximately 1,000 high school students on mathematics, and found that those with access to unrestricted use of Generative AI, scored 17% worse on the subsequent unassisted examinations, compared to students who never had access to Generative AI. “Generative AI without guardrails can harm learning” – PNAS An important thing to note is that the students who used AI, believed that AI had actually helped them with learning. The authors described this fallacy as an illusion of competence. Similarly, a Brazilian experiment by André Barcauj, found that students who learned with AI assistance retained significantly less knowledge 45 days later, congruent with the desirable difficulties framework in Cognitive Science, which highlights that effortful processing is an integral part of building a durable memory. “ChatGPT as a Cognitive Crutch” – International Journal of Educational Research Open
However, a pertinent point that needs to be highlighted is that these trends cannot be applied universally across Indian classrooms. The cognitive analysis documented in the research cited above, was observed among students in well resourced university settings. However, India, even before it can tackle the issues propping up with the usage of AI, has a more fundamental problem. The Annual Status of Education Report, 2024, found that only 23.4% of Class III students in government schools could read a Class II level text, and barely 30.7% of Class V students could perform basic division. ASER 2024 – Pratham Foundation Adding to the same, the Unified District Information System for Education Plus 2024-25 data showcases that although 63.5% of schools nationally now have internet access, Government schools fall behind private schools at 58.6% compared to 77.1%, the gap broadens in states like Bihar, where only 23.8% of the schools have functional computers. UDISE+ 2024–25 – Ministry of Education The reason for highlighting these statistics is to emphasize that where foundational literacy itself remains unresolved, the integration of AI tools without any guidelines, principles, or guardrails, risks exacerbating the existing learning deficit.
The above data is particularly consequential for India when compared to its AI usage. The country has the largest ChatGPT user base globally, and evidence from OpenAI highlights that users aged between 18 to 24 account for nearly half of all the prompts by Indians to ChatGPT, while users under the age of 30 account for 80%. TechCrunch report on OpenAI India usage data On the sidelines, OpenAI announced a partnership with the Indian Institute of Technology, Delhi, the Indian Institute of Management, Ahmedabad, and the All India Institute of Medical Sciences, New Delhi, providing ChatGPT’s enterprise access to over 1 lakh students. Business Standard – OpenAI partnerships with Indian institutions Additionally, The All India Council for Technical Education (AICTE) declared 2025 as the Year of AI, supporting the same by launching dedicated programs across 14,000 institutions, which admit approximately 40 lakh students. AICTE Year of AI announcement This is indicative of the pace of AI adoption in the education sector.
Now, while AI in the education sector has its own merits and should also be looked at independently, particularly from the perspective of how it helps in augmenting the education sector, the present literature suggests that this needs to be complemented with a set of guardrails in place. It would be unfair to conclude that AI in the education sector is inherently harmful, as even the Wharton study cited above found that students using a pedagogically scaffolded AI tutor showed no learning deficit at all.
The Economic Survey itself puts out the strongest case. It cites The India Skills Report 2025 by Wheebox, which argues that as the momentum for automation of routine tasks increases, employers look for a workforce that is proficient in foundational skills such as problem solving, rather than technical specialisation. India Skills Report 2025 – Wheebox Hence, the core issue is that if unguided AI dents the very foundational cognitive skills that employers demand, India is simultaneously accelerating AI adoption and undermining the human capabilities that its economy strongly demands.
As of 2026, the University Grants Commission’s (UGC) Plagiarism Regulations, which were enacted in 2018, remain the core framework for maintaining the integrity of research outputs. UGC Plagiarism Regulations, 2018 These regulations, however, have no clauses specifically addressing AI generated content. In practice though, the commission has begun enforcement against AI generated content. For instance, UGC rejected several PhD theses from Babasaheb Bhimrao Ambedkar Bihar University in Muzaffarpur after it found over 40% AI written material. News9Live – UGC rejection of PhD theses Although a key point to be noted here is that this investigation was based on Turnitin detection, and not because of any prevailing guidelines or procedures to undertake the same. Individual universities have been left to develop their own policies. AICTE, on the other hand, has been comparatively proactive. It has proposed amendments to its plagiarism rules to include AI generated text and has mandated institutions to submit their AI implementation plans. AICTE PhD rules update
Globally, there are a couple of examples that provide a guiding path. For instance, Australia’s Tertiary Education Quality and Standards Agency required all its higher education institutions to submit institutional AI action plans by mid-2024 and has since then published a detailed framework for assessment reform. TEQSA Gen AI strategies and assessment reform Similarly, The United Kingdom’s Russell Group of 24 leading Universities published and updated principles on Generative AI in education. Russell Group AI principles The EU AI Act considers education as a high-risk domain. EU AI Act 2024 In relation to this, while India’s Digital Personal Data Protection Act of 2023 does govern how student data is collected and processed, it does not regulate design or deployment of AI systems in education. Digital Personal Data Protection Act, 2023
Authored by Raunaq Sharma, Senior Research Associate, The Dialogue.