If you work in or around Indian financial services and AI, the RBI FREE-AI framework is the document that landed on your desk in August 2025 and is not going away. It is not a binding regulation, and that is precisely why a lot of people have skimmed it and moved on. That is a mistake. Frameworks like this are how the Reserve Bank signals where binding rules are heading, and getting ahead of it is far cheaper than catching up later. It also sits alongside the broader responsible AI principles India has been building since 2018.
Here is a clear-eyed walk through what the RBI FREE-AI framework actually is, stripped of the acronym soup.
Where the RBI FREE-AI framework came from
In December 2024 the Reserve Bank of India set up a committee to work out a framework for responsible and ethical AI in the financial sector. It was chaired by Professor Pushpak Bhattacharyya of IIT Bombay, and it delivered its report, titled Framework for Responsible and Ethical Enablement of Artificial Intelligence, which shortens to FREE-AI, in August 2025.
The name is doing real work. The word that matters most in it is enablement. This is not a framework written to restrain AI in finance. It is written to encourage its adoption while keeping it responsible. That framing runs through the whole document, and it is why the tone is closer to a roadmap than a rulebook.
The committee also grounded the work in reality by surveying regulated entities. The finding worth remembering is the gap it exposed: only around a fifth of institutions were actually deploying AI, while roughly two thirds said they were interested. The framework is aimed squarely at closing that distance between intent and action.
The seven Sutras
At the heart of the framework sit seven guiding principles, which the report calls Sutras. They are meant to be the values that anchor every specific recommendation. In plain terms they run roughly like this: trust is the foundation of everything; people should be at the centre, with humans overseeing consequential decisions; innovation carries responsibility alongside it; there must be fairness and equity, with bias actively managed; there should be accountability, so someone owns each outcome; systems should be understandable rather than inscrutable black boxes; and safety, resilience and sustainability should be built in rather than bolted on.
If those sound like principles you have seen in other responsible-AI documents, that is the point. The Sutras are deliberately not novel. They are a shared vocabulary so that everything downstream has something stable to hang from.
Six pillars and 26 recommendations
Below the principles, the RBI FREE-AI framework gets concrete. It lays out 26 recommendations organised under six pillars, and the six pillars are themselves grouped into two halves that capture the framework’s whole philosophy.
The first half is about enabling innovation, and it covers three pillars: infrastructure, policy, and capacity. The thinking here is that you cannot ask institutions to adopt AI responsibly if the groundwork is missing. So the framework pushes for shared infrastructure, supportive policy, and investment in skills across the sector.
The second half is about mitigating risk, and it also covers three pillars: governance, protection, and assurance. This is where the responsibility side lives. It means governance structures so AI is overseen properly, protection for consumers and their data, and assurance mechanisms so that claims about AI systems can actually be checked.
The symmetry is intentional. Three pillars to help you move forward, three to keep you safe while you do. The framework’s core argument is that these are not in tension. You get durable adoption only when enablement and safeguards advance together.
What it means if you are a regulated entity
The framework is advisory, not binding. Nobody is going to penalise you tomorrow for not implementing recommendation 14. But treating advisory as ignorable misreads how regulation tends to work in Indian finance. The direction of travel here is unusually clear, and the sensible institutions are reading it as an early sight of where supervisory expectations are heading.
Practically, the useful move is not to try to implement all 26 recommendations at once. It is to look at the two halves and ask where you are weakest. Most institutions we speak to are further along on the enablement side, the appetite and even the tooling. They are weaker on the risk side: the governance, the documentation, the assurance that would let them prove an AI system behaves as claimed. That asymmetry is exactly what the framework is nudging the sector to correct.
The honest summary
The RBI FREE-AI framework is a signal, not a statute. It tells you that the Reserve Bank wants Indian financial institutions to adopt AI, and to do it responsibly. It has laid out a reasonably practical map of what responsible looks like across seven principles, six pillars and 26 recommendations. Reading it as a compliance chore misses the point. Reading it as a preview of the questions your supervisor will eventually ask is much closer to the mark.
At humaineeti a lot of our work with financial-sector clients sits exactly on the risk-mitigation side of this framework. We build the governance, documentation and assurance that lets an institution deploy AI and still answer for it. If you are trying to translate FREE-AI from principles into something your teams can actually operate, that gap between the framework and day-to-day practice is where the real work is.
Comments 1