India proposes that AI models travel to the data, not the other way round
A draft handbook from India's technology ministry sets out how data owners could let AI be trained without handing their data over.
India's technology ministry has published a draft handbook on a framework it calls DEPA for AI Training. Tribune India reports that it was launched on 31 October 2026 in Bengaluru. The government's Principal Scientific Adviser summed it up as the model travelling to the data, rather than the data travelling to the model.
As The Register describes it, hospitals, banks and government agencies would act as training data providers. They share anonymised, encrypted data through exchange platforms, and those building models train them there, under rules the providers set. A separate secure environment gives a neutral place to run the work. The handbook also covers provenance, so the origin, permissions and use of data can be tracked during training.
It is labelled a draft, and many details are still open. For people who own data, the direction is plain: a government is writing down a route where data is used without being handed over. Set terms, a tracked trail and a neutral place to compute are what could make that route trusted.
