MeitY launches DEPA handbook so AI can train on data without the data leaving home

The Ministry of Electronics and Information Technology (MeitY) has launched a Handbook on Data Empowerment and Protection Architecture (DEPA) for AI Training. MeitY Secretary S. Krishnan unveiled it at a roundtable in Bengaluru on 31 October 2026, with support from the iSPIRT Foundation, according to the government's own press release.

In simple words, the handbook is a rulebook for sharing data to train AI without handing that data over.

The problem it tries to solve

Good AI models need a lot of good data. In India, much of that data sits with banks, hospitals, research labs and government bodies. They often don't share it, and for real reasons: customer privacy, business secrets, intellectual property, and the fear of losing control once the data leaves their systems.

So a startup building, say, a credit model or a medical AI tool may never get access to the data that would make its product work.

How DEPA Training works

The press release describes DEPA Training as a "techno-legal" framework. That means it mixes legal rules with technical controls:

• Contracts and permissions: data can be used only for an agreed purpose.
• Controlled computing environments: the training happens in a locked-down setup, not on the AI builder's own machines.
• Privacy safeguards: technical tools that protect personal and confidential information.
• Data provenance: a record of where each dataset came from, what permission it carried, and how it was used during training.

The institution that owns the data keeps control of it throughout. Principal Scientific Adviser Prof. Ajay Kumar Sood summed up the idea at the event: "the model travels to the data, rather than the data travelling to the model."

Who was in the room

Besides MeitY and the Principal Scientific Adviser, the release names speakers from iSPIRT, HDFC Bank, Infosys co-founder Kris Gopalakrishnan (as chairman of Axilor Ventures), Emergent AI, Sarvam and YourStory. HDFC Bank's representative said the bank has already worked with DEPA Inference, an earlier part of the same approach, and sees value in extending it to training.

iSPIRT co-founder Sharad Sharma compared the moment to the launch of UPI in 2016. That is his view; how widely the framework is adopted will decide whether the comparison holds.

What we don't know yet

The press release does not give a timeline, say whether using the framework will be voluntary or required, or name the first datasets or sectors that will use it. We will update this story if MeitY shares those details.

Day1 take

For AI startups, data access is often a bigger wall than money or compute. If banks, hospitals and labs start opening their data through DEPA-style setups, smaller Indian teams could train models on data they could never buy. Founders in fintech, health and research AI should read the handbook now and ask early partners whether they would join a pilot.

Sources: PIB press release, MeitY, "MeitY Launches DEPA for AI Training Handbook to Advance Trusted Data Collaboration for AI" (primary, 31 Oct 2026). Quotes are taken from that release. Lead via Fortune India and YourStory reports (31 Oct 2026).