Responsible AI for All (NITI Aayog, 2021)

Short answer

NITI Aayog published two approach papers. Part 1, "Principles for Responsible AI", is dated February 2021 (cover, PDF page 1). Part 2, "Operationalizing Principles for Responsible AI", is dated August 2021 (cover, PDF page 1). Both are policy papers.

Law: none. These papers do not bind you.

What it means for you

If you build AI, these principles are a useful checklist. Later papers repeat many of the same ideas. The AI Governance Guidelines, 2025 use seven "sutras" of their own. Following them is good practice. It is not a legal duty.

How it works

Seven principles (Part 1, PDF page 13):

1. Safety and reliability.
2. Equality.
3. Inclusivity and non-discrimination.
4. Privacy and security.
5. Transparency.
6. Accountability.
7. Protection and reinforcement of positive human values.

Part 1 also looks at two kinds of effect. "Systems considerations" are direct effects of an AI system. Its example is AI cancer screening, which must protect patient privacy in its design (PDF page 18).

Part 2: putting it into practice. Part 2 has a section on the role of government and a section on actions for the private sector and research bodies (contents, PDF page 8). On regulation, it recommends a risk-based approach. Rules should match how likely and how serious a harm is (PDF page 12).

Good practice drawn from the principles:

• Test your system for errors and unfair results before launch.
• Tell users when AI is making or shaping a decision about them.
• Name a person who is responsible for each AI system.
• Protect personal data. Here the DPDP Act, 2023 does set legal duties.

Where it fits

These papers sit between the 2018 strategy and the 2025 Guidelines. The 2018 strategy asked what AI could do for India. These papers ask how AI should behave. The 2025 Guidelines then turn similar ideas into advice for companies and government. None of them is a law. If a regulator such as RBI or SEBI sets AI rules for its sector, those rules can bind the firms it regulates. Check the sector pages for those.

Good practice: write a short note for each AI system. Say which of the seven principles matter most for it, and how you check them. Review the note when the system changes. Share it with the people who build, test and sell the system, so everyone works to the same standard.

What's still open

• Neither paper sets penalties or a regulator. No law has adopted these seven principles as written.

See also the AI Governance Guidelines, 2025.