Responsible AI for All: Approach Document for India, Part 1 (Principles, February 2021) and Part 2 (Operationalizing the Principles, August 2021)

NITI Aayog's 2018 National Strategy for Artificial Intelligence devoted a chapter to ethics, privacy and security and recommended ethics councils at research centres, but it did not set out a set of operative principles.² A follow-up was started in 2019, when NITI Aayog began stakeholder consultations with the World Economic Forum on how to approach the responsible use of emerging technologies. According to NITI Aayog's own later account, this work "culminated in 2021, with the release of a two-part approach paper" that identified principles for responsible design, development and deployment of AI in India and set out enforcement mechanisms for operationalising them.¹

The documents belong to the series 'Responsible AI #AIforAll'. Part 1 appeared in February 2021 and Part 2 in August 2021. The third paper in the series, on facial recognition technology, was released for comment in November 2022.¹ The original PDFs of Parts 1 and 2 could not be retrieved when this entry was prepared, so the account here relies on NITI Aayog's own later description of them in the 2022 paper, on secondary summaries, and on the links to the originals that appear in that paper.¹³⁵⁶

The 2022 NITI Aayog paper states that the principles "have been developed by first identifying systemic considerations prevalent among AI systems across the world, and identifying principles that may be used to mitigate the identified considerations."¹ Its evaluation table paraphrases each principle as follows:¹

Wikipedia's summary adds that Part 1 treats both system-level issues, such as decision-making and accountability, and societal issues, such as the effects of automation on employment.³ The principles are explicitly rooted in "constitutional morality": the 2022 paper describes this as compliance with the "constitutional ethos", and says that India's AI ethics principles were envisioned with constitutional morality as their cornerstone.¹ That framing ties AI governance to constitutional rights, in particular the right to privacy under Article 21 as recognised in Justice K.S. Puttaswamy v. Union of India, which the 2022 paper uses as its legal yardstick (legality, reasonability and proportionality) for state use of facial recognition.¹

Part 2, titled 'Operationalizing Principles for Responsible AI', was published in August 2021. The 2022 paper describes it as setting out "enforcement mechanisms" for the principles, and Wikipedia's summary lists its subject matter as regulatory and policy interventions, capacity building and incentives for ethical practice.¹³

One mechanism that NITI Aayog's later papers attribute to the approach document is the ethics committee. The 2022 facial recognition paper says that the responsible AI approach paper "recommends that organisations deploying an AI system can constitute an ethical committee to assess the ethical implications and oversee mitigation measures". It then applies that idea to facial recognition, asking procuring agencies to set up autonomous committees that draft explainability and fairness standards, supervise compliance with standard operating procedures, define grievance redress and publish annual reports, and asking vendors to set up separate internal ethics committees.¹ The 2022 paper also pairs each principle with a 'systems failure' consideration, so that an assessment asks which failure mode a principle is meant to mitigate and what measures exist against it.¹

The approach documents also feed into government procurement. The 2022 paper recommends that requests for proposals for AI systems be "in line with government strategy papers such as the National Strategy for AI, 2018 and the Responsible AI 2021 papers".¹

The principles are guidance. They create no obligations for private firms, set up no regulator, and have no penalties. Their main use so far has been inside government, where NITI Aayog applied them to a specific use case. In its 2022 facial recognition paper, prepared with the Vidhi Centre for Legal Policy as knowledge partner, it assessed the DigiYatra biometric boarding programme against each of the seven principles and made detailed recommendations, which were open for public comment until 30 November 2022.¹⁷

Later official policy has drawn on the 2021 work in more general ways. MediaNama's explainer on the November 2025 India AI Governance Guidelines lists 'NITI Aayog (Feb, 2021): Responsible AI for All' among the earlier policy documents.⁴ The 2025 Guidelines themselves, however, adopt seven new sutras (Trust is the Foundation, People First, Innovation over Restraint, Fairness and Equity, Accountability, Understandable by Design, and Safety, Resilience and Sustainability), which they say were adapted from the Reserve Bank of India's FREE-AI Committee report.⁸ Several ideas overlap with NITI Aayog's 2021 list: safety, fairness and non-discrimination, accountability and transparency. The 2025 text adds two principles that the 2021 list does not contain, trust as the foundation and an explicit priority for innovation over restraint.⁸

The committee that drafted the 2025 guidelines included Debjani Ghosh, a Distinguished Fellow at NITI Aayog, and Amlan Mohanty, a NITI Aayog non-resident fellow who was the lead writer.⁸

The 2021 documents were among the first official Indian statements of AI ethics principles, and they placed the Constitution, not a foreign framework, at the base of Indian AI governance. Because they set out principles and mechanisms and not rules, their effect depends on voluntary uptake. The same 2022 paper that tested them says the absence of a data protection law left a gap in the legal framework for facial recognition and that transparency and bias are non-privacy risks needing "separate regulation, either through codes of practice, industry manuals and self-regulation, or through more formal modes like statute".¹

India has since chosen not to enact a standalone AI statute. The 2025 India AI Governance Guidelines conclude that "a separate law to regulate AI is not needed given the current assessment of risks" and favour existing laws, voluntary frameworks and techno-legal tools, with some baseline measures to be made mandatory later by sectoral regulators.⁸ That is the same combination of soft principles and sectoral enforcement that the 2021 documents began.

The principles' first applied test drew criticism from civil society. In comments on the 2022 facial recognition draft, the Internet Freedom Foundation called for a blanket ban on law-enforcement use of facial recognition and on emotion recognition, and argued that the paper's explainability recommendations were unrealistic; the Software Freedom Law Centre asked for 'informed explicit consent' and for a separate AI law.⁹¹⁰ Those comments are covered in the entry on the 2022 paper.