India proposes DEPA framework to let AI models train on sensitive data without it leaving the country
2026-11-02
India's Ministry of Electronics and Information Technology proposed the Data Empowerment and Protection Architecture (DEPA) for AI Training, under which anonymized, encrypted data from hospitals, banks, and government agencies would be shared through government-run "Data Exchange Platforms" that AI model trainers access inside hardware-secured environments, so that, as described by the government's Principal Scientific Adviser, "the model travels to the data, rather than the data travelling to the model."
Significance 3: A national government framework proposal for sovereign AI data governance is a credible policy signal shaping how regulated health and financial data can be used to train AI in a major market, though it remains a proposal, not yet implemented.
Implications · machine-drafted, not owner judgment
If implemented, DEPA could make India a distinct data-governance zone: training runs touching Indian health or financial data would need to happen inside government-controlled secure environments rather than a lab's usual pipeline, raising the cost of serving India's market for foreign frontier labs and potentially favoring labs or local players willing to build India-specific infrastructure. Other large regulated markets (EU, Gulf states) could borrow the same template, extending the geography/policy fragmentation already being tracked elsewhere (the US DeepSeek procurement ban, China's export pre-approval proposal).
- Any frontier lab (OpenAI, Google, Anthropic) committing to build inside a DEPA Data Exchange Platform within the next year
- DEPA moving from framework document to an operating pilot