Long before the sector regulators started writing their own AI rulebooks, NITI Aayog did something quieter and more foundational. In its Responsible AI approach documents, the government’s policy think tank set out seven principles for what responsible AI in India should mean. They are not law. They are something arguably more useful: the shared vocabulary that later, more specific rules keep echoing.
If you only ever read one thing on Indian AI governance, these principles are a better starting point than any single regulation, because they explain the intent that everything else is trying to operationalise.
Where these principles come from
The groundwork was the National Strategy for Artificial Intelligence in 2018, with its memorable AIforAll framing. NITI Aayog then followed up with the Responsible AI approach documents. The first part, in early 2021, laid out the principles. The second part, later that year, focused on how to actually put them into practice. That two-step, principles then operationalisation, tells you the intent was never to produce a nice list and stop.
The seven responsible AI India principles are usefully split into two groups. Some concern the AI system itself. Others concern its effect on society. Keeping that split in mind makes the whole set easier to reason about.
The system-level principles
The first group is about how the system is built and behaves.
Safety and reliability: the system should do what it is supposed to, behave predictably, and not cause harm through malfunction. This is the baseline. A system that is unsafe or unreliable fails before any of the subtler concerns even matter.
Equality: like cases should be treated alike. People in similar circumstances should get similar outcomes from the system, rather than arbitrary variation.
Inclusivity and non-discrimination: the system should not encode or amplify bias against groups, and it should work across the genuine diversity of the people it touches. In a country as varied as India, this is not a box-tick. It is one of the hardest and most important requirements.
Privacy and security: people’s data should be protected, both in how it is used and how it is kept safe. This principle predates and dovetails with the data protection regime that has since taken shape.
The society-level principles
The second group looks outward, at the relationship between the system and the people and institutions around it.
Transparency: it should be possible to understand, at an appropriate level, how the system works and how it reaches its decisions. Not every user needs the maths, but the system should not be an unaccountable black box.
Accountability: there must be someone answerable for what the system does. Responsibility cannot evaporate into the software. A human or an organisation owns the outcomes.
Protection and reinforcement of positive human values: the system should support, rather than erode, the values a society wants to keep. This is the most abstract of the seven, and it is the one that insists technology serve human ends rather than the other way round.
Why these responsible AI India principles still matter
It would be easy to treat a set of principles from 2021 as dated. That misreads their role. Principles like these do not go stale the way a specific technical rule does, because they describe intent rather than mechanism. When a sector regulator later writes something concrete, whether in finance, such as the RBI FREE-AI framework, health, or elsewhere, you can usually trace its clauses straight back to one or more of these seven. Understanding the principles helps you anticipate the specifics.
For anyone building AI systems for the Indian market, the practical value is as a checklist of the questions you will eventually have to answer. Can you show your system is safe and reliable? Can you demonstrate it does not discriminate across the populations it serves? Is it transparent enough to explain? Is someone genuinely accountable for it? Get those answers ready early, because the direction of Indian AI governance guarantees they will be asked.
From principles to practice
The gap, as always, is between a principle everyone nods along to and a system that actually embodies it. Non-discrimination is a fine word until you have to test a model across dozens of demographic slices. Accountability sounds obvious until you try to reconstruct why an automated decision was made six months ago.
This is exactly the translation work we focus on at humaineeti: taking responsible AI in India from a set of principles everyone agrees with into governance, testing and documentation that a real system can be held to. The principles tell you what good looks like. Closing the distance to a system that genuinely meets them is where the effort lives.
Comments 1