MeitY unveils DEPA handbook to enable trusted data sharing for AI training

Ministry of Electronics and Information Technology (MeitY) has launched a handbook on Data Empowerment and Protection Architecture (DEPA) for AI training, an official release said Saturday.

The handbook outlines a techno-legal framework that allows institutions to collaborate on data for AI training while keeping control of their data.

IT Secretary S Krishnan unveiled the handbook, which was developed with support from the iSPIRT Foundation, at a roundtable in Bengaluru.

"The framework combines contractual permissions, controlled computing environments and privacy-enhancing safeguards. It also incorporates data provenance, enabling the origin, permissions and use of data to be traced throughout the training process, supporting accountability and transparency," the release said.

As AI becomes central to economic and social development, access to diverse, high-quality data is critical.

Yet concerns around privacy, confidentiality, intellectual property and institutional control often limit data sharing, the release said adding DEPA training addresses these challenges.

"India's AI ambitions must be matched by the ability to use data responsibly and securely," Krishnan said, adding that MeitY is committed to taking the framework forward.

Principal Scientific Adviser Ajay Kumar Sood explained that under DEPA, the model travels to the data rather than the reverse.

He said this enables multi-party and cross-border collaboration, and described DEPA as India's contribution to global science diplomacy.

The roundtable drew representatives from government, industry, academia, research among other stakeholders.

Framework to protect sensitive data

The framework is aimed at addressing a practical problem that has become more pressing with the growth of generative AI--organisations may have access to valuable datasets but are often reluctant to move or share the underlying data. This is particularly relevant for sectors such as healthcare, finance, education and public services, where datasets can contain sensitive or commercially valuable information.

Under the DEPA approach, organisations can retain control over their data while allowing AI models or approved computing processes to work within controlled environments. This could make it easier for multiple institutions to contribute data to a common AI training exercise without creating a central pool containing all the underlying datasets. The provenance layer is also intended to provide a record of where the data came from, what permissions apply to it and how it was used. This can help institutions establish clearer accountability as AI systems are developed and trained.