Policy Update
Ashi Verma
Background
In the era of rapid digital transformation, Artificial Intelligence has emerged as a critical driver of economic growth and governance in India. However, its unregulated expansion has raised serious concerns regarding ethical use, data sovereignty, and accountability. To address this regulatory gap, the Government of India launched the MANAV Framework in 2026. The framework was formally unveiled by Prime Minister Narendra Modi during the India AI Impact Summit, 2026 at Bharat Mandapam, New Delhi. It has been conceptualized by the Ministry of Electronics and Information Technology (MeitY) under the IndiaAI Mission, following the recommendations of a drafting committee constituted in July 2025.
MANAV is an acronym for Moral, Accountable, National Sovereignty, Accessible and Valid. Symbolically, the word ‘MANAV’ translates to ‘human’, which reflects its core philosophy of human-centric AI. Grounded in seven foundational Sutras, the framework adopts a techno-legal approach that seeks to balance technological innovation with the imperative of safe, trusted and inclusive AI governance.
Functioning
- Three-Tier Institutional Architecture: The framework functions through a whole of government institutional model. At the top is the AI Governance Group, chaired by MeitY, which acts as the apex policy-making body. Supporting it is the Technology and Policy Expert Committee comprising industry experts, academicians, and civil society to draft technical standards and ethical codes. At the execution level, the AI Safety Institute under Indian AI mission functions as a risk auditing body, responsible for testing AI models for bias, safety, and deepfake threats before public deployment.
- Five-Layered IndiaAI Stack for Implementation: Operationally, it functions via an integrated stack. The Data Layer (IndiaAI Kosh) provides more than 15k anonymized datasets for training. The Compute Layer (IndiaAI Compute Portal) offers 38k-45k+ GPUs at subsidized rates to startups by mid-2026. The Model Layer promotes indigenous foundation models. The Application Layer deploys AI in citizen services
- Principle-Based Functioning through Seven Sutras: People First Innovation over Restraint, Fairness & Equity Accountability, Understandable by Design, Safety, Resilience & Sustainability. Every AI application used by the government must undergo mandatory bias testing, ensure human-in-the-loop oversight for critical decisions, and label all AI-generated content to prevent misinformation.
- Techno-Legal Enforcement Model: Unlike the EU’s hard law approach, MANAV functions on a techno-legal model. It does not create a new standalone Act initially. Instead, it enforces its principles through existing legal instruments like the IT Act, 2000, Digital Personal Data Protection (DPDP) Act, 2023, and IT Rules, 2026. This makes developers and platforms legally accountable while retaining flexibility for innovation.
Performance
The MANAV Framework has demonstrated early but significant performance in shaping India’s AI governance architecture.
Firstly, in the domain of AI safety and ethics, the framework has institutionalized oversight through the AI Safety Institute. Since its inception, the Institute has been auditing Large Language Models, including indigenous models like BharatGen and Sarvam, for bias, toxicity, and deepfake risks. The mandatory AI-content labelling provision has also shown early positive impact in curbing AI-generated misinformation during elections.
Secondly, the framework has performed effectively in democratizing critical AI resources. Through its integration with the IndiaAI Mission, the IndiaAI Kosh (AIKosha), launched in March 2025, has significantly expanded its repository from just over 300 datasets at launch to over 14,000+ datasets and 300+ AI models by mid 2026, while the IndiaAI Compute Portal has provisioned over 38,000 GPUs to startups and academic institutions. This has reduced dependency on foreign cloud providers and enabled the development of indigenous foundation models.
Thirdly, on the parameter of inclusive and citizen-centric application, the ‘A for Accessible and Inclusive’ principle of MANAV has been successfully operationalized. Initiatives such as Bhashini for translation of judicial and governmental documents in Indian languages, AI-based claim verification under PMJAY, and Kisan e-Mitra advisories for farmers, reflect its on-ground impact.
However, the framework’s performance is constrained by structural limitations. It lacks a standalone statutory authority, unlike the EU AI Act, and relies on a techno-legal enforcement model based on the IT Act, 2000 and the DPDP Act, 2023, which results in enforcement ambiguity. Furthermore, the auditing capacity of the AI Safety Institute is inadequate relative to the rapid pace of model development, and the participation of Tier-2 and Tier-3 city startups in GPU allocation remains below 25%, highlighting a regional digital divide that requires urgent policy intervention.
Impact
The MANAV Framework has had a transformative impact on India’s AI governance ecosystem. Its primary impact lies in establishing a robust mechanism for AI safety and accountability. Through the AI Safety Institute and mandatory AI-content labelling, it has institutionalized the auditing of Large Language Models for bias and toxicity and has significantly curbed AI-generated misinformation, thereby strengthening digital trust and electoral integrity.
Secondly, it has democratized AI development by making critical infrastructure publicly accessible. The provision of over 350 datasets via IndiaAI Kosh and 14,000 GPUs via the Compute Portal to more than 800 startups and academic institutions has reduced dependence on foreign cloud providers and enabled the creation of indigenous foundation models in all 22 scheduled languages, promoting both Atmanirbhar Bharat and linguistic inclusion.
Thirdly, its citizen-centric impact is evident in the last-mile delivery of AI for public good. Applications such as Bhashini for multilingual governance, AI-enabled claim verification under PMJAY, and Kisan e-Mitra advisories for over 2 crore farmers have translated AI from a frontier technology into an instrument of inclusive development and social welfare.
Emerging Issues
The MANAV framework, while being a progressive step, faces several emerging structural and operational challenges
- Absence of Statutory Backing: The framework does not have a dedicated legislation like the EU AI Act. It relies on existing laws such as the IT Act, 2000 and DPDP Act, 2023. This techno-legal model creates enforcement ambiguity and jurisdictional overlap between MeitY, CERT-In and the newly formed AI Safety Institute.
- Institutional Capacity Constraint: The pace of AI model development is far faster than the auditing capacity of the AI Safety Institute. With over 23 LLMs audited in 7 months, the Institute is overburdened. Moreover, there is a lack of standardized, India-specific benchmarks for auditing bias, toxicity and hallucination in indigenous models.
- Regional Digital Divide: The benefits of IndiaAI Compute and IndiaAI Kosh remain concentrated. Data indicates that less than 25% of GPU allocation has reached startups in Tier-2 and Tier-3 cities. This concentration risks widening the existing digital and innovation divide.
- Innovation-Regulation Dilemma: The mandatory compliance requirements, such as AI-content labelling and self-declaration, increase compliance costs for startups. This may lead to regulatory overreach and deter early-stage innovation, or push developers towards regulatory arbitrage.
- Data and Accountability Vacuum: While IndiaAI Kosh provides 350+ datasets, there is no robust mechanism to prevent re-identification or ensure data sovereignty. Furthermore, the framework lacks a clear liability and accountability structure to fix responsibility in case of AI-induced harms in sensitive sectors like healthcare and agriculture.
- Limited Participatory Governance: The framework was largely a top-down initiative. Participation of civil society, academia from non-metro institutions, and end-users in its design and feedback loop remains limited, which affects its inclusive legitimacy.
Way Forward
To ensure the long-term effectiveness of the MANAV Framework, it needs to transition from a set of guidelines to a robust, legally empowered ecosystem. Firstly, there is a need to enact a dedicated and risk-based AI legislation that provides statutory powers to the AI Safety Institute, clearly defines high-risk AI applications, and establishes a proportionate liability and accountability framework. Secondly, the institutional capacity of the AI Safety Institute must be augmented through public-private partnerships, creation of India-specific auditing benchmarks, and establishment of regional AI safety hubs to bridge the gap between innovation hubs and Tier 2 and Tier 3 cities.
This should be complemented by a more inclusive GPU allocation and skilling policy under the IndiaAI Mission. Finally, the framework must adopt a participatory and adaptive governance model by institutionalizing multi-stakeholder consultations involving civil society, academia and startups, launching mass awareness campaigns on AI literacy and content labelling, and creating a periodic review mechanism to keep the framework aligned with the rapidly evolving global AI landscape.
References
1. Prime Minister’s Office. (2026, February 19). English rendering of PM’s address at India AI Impact Summit 2026 MANAV Vision. Press Information Bureau. https://pib.gov.in/PressReleasePage.aspx?PRID=2104421
2. Press Information Bureau. (2026, February 19). India AI Impact Summit 2026 – PM unveils MANAV vision.
https://pib.gov.in/PressReleasePage.aspx?PRID=2104402
3. Ministry of Electronics and Information Technology. (2025, March 6). The government launched AIKosha and IndiaAI Compute Portal with 14,000 GPUs. PIB. https://pib.gov.in/PressReleasePage.aspx?PRID=2108738
4. IndiaAI. (2025, October 7). Scaling up efforts for deepfake detection, bias mitigation and AI Safety Institute.
5. Ministry of Electronics and Information Technology. (2025, November). India AI Governance Guidelines.
ttps://www.meity.gov.in/writereaddata/files/India-AI-Governance-Guidelines.pdf
6. NITI Aayog. (2021). National Strategy for Artificial Intelligence Government of India. https://www.niti.gov.in/sites/default/files/2023-03/National-Strategy-for-Artificial-Intelligence.pdf
7. Ministry of Electronics and Information Technology. (2023). Digital Personal Data Protection Act, 2023.
https://www.meity.gov.in/writereaddata/files/Digital%20Personal%20Data%20Protection%20Act%202023.pdf
About the Contributor
Ashi is pursuing B.A Program (History and Political Science) from Maitreyi College (NCWEB) South Campus Delhi University with a major in Political Science. She is passionate about governance, public policy, social issues and digital innovation, with a keen interest in research, academic writing, and data documentation. Through research and policy analysis, she aims to contribute to evidence-based policymaking and create meaningful social impact.
Acknowledgement
I would like to express my sincere gratitude to IMPRI (Impact and Policy Research Institute) for giving me the opportunity to work as an intern and write on such an important topic. I would also like to acknowledge the official MeitY and IndiaAI Mission for making authentic data available on the PIB Portal that helped me to complete this article.
Disclaimer
The views and opinions expressed in this article are those of the author and do not necessarily reflect the official policy or position of IMPRI (Impact and Policy Research Institute) or any other affiliated organisation. The information presented in this article is based on publicly available government sources and is intended solely for academic and research purposes
Name of the Reviewers:
Arjya, Kaustav Majumdar
Read more at IMPRI:
Health Security at the Heart of Viksit Odisha
A Scheme for Promotion of Innovation, Rural Industries and Entrepreneurship (ASPIRE), 2015
