India launches DEPA framework for governed AI training data

India’s Ministry of Electronics and Information Technology launched a Data Empowerment and Protection Architecture handbook for AI training on October 31. The handbook sets out rules for models to access institutional data in controlled environments while participating organizations retain control of their datasets.

The ministry describes DEPA for AI Training as a techno-legal framework for purpose-bound data collaboration. It brings together contractual permissions, controlled computing environments and privacy-enhancing safeguards intended to govern the use of data for model training.

Ajay Kumar Sood, India’s principal scientific adviser, described the model as one in which “the model travels to the data, rather than the data travelling to the model.” According to the ministry, the framework also incorporates data provenance so that a dataset’s origin, permissions and use can be traced through the training process.

The launch covers a handbook and governance framework, not an announced operational rollout. The ministry’s release did not identify a live secure environment, a pilot, committed data providers or approved model developers, and it gave no deployment timetable.

MeitY Secretary S. Krishnan said the framework could enable trusted data collaboration at scale while protecting privacy, commercial confidentiality and institutional control. Those are ministry claims, not demonstrated results. Krishnan unveiled the handbook at a Bengaluru roundtable with support from the iSPIRT Foundation.