Discussion

India proposes a ‘model goes to the data’ framework for AI training

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#10559

India is proposing a framework for training AI models on locally held data without routinely moving that data to model-makers. The plan could give hospitals, banks and government agencies a way to share data under rules they set, though the report says it is unclear how much work remains before the system is operational.

Watch Desk analysis

What happened

India’s Ministry of Electronics and Information Technology calls the proposal the Data Empowerment and Protection Architecture for AI Training, or DEPA. The Register reports that government science adviser Ajay Kumar Sood described its principle as “the model travels to the data, rather than the data travelling to the model”.

Under the framework, data providers would share anonymised, encrypted information through a Data Exchange Platform. Model developers would access it inside hardware-based Secure Data Environments, where the framework says data would remain encrypted and isolated during processing. Providers could set rules for how their data is used; the framework also calls for differential privacy and immutable audit logs. Read The Register’s report.

Why it matters

The proposal tries to make data useful for AI training without making unrestricted data transfer the price of admission. That matters in fields such as health and finance, where sensitive records may support research or services but are not ordinary training fodder.

The framework describes a proposed architecture, not proof that its protections have been tested in practice or that a working system is ready. The Register says India already operates several data exchanges, but reports that the remaining work to make DEPA a reality is unclear.

Our read

“Bring the model to the data” is a compelling answer to a real governance problem, provided the safeguards work as described and data providers retain meaningful control. The useful test is not the elegance of the diagram; it is whether organisations can use the system without exposing records or losing oversight.

What to watch

  • Whether India publishes implementation plans, timelines or technical standards for DEPA.
  • How access rules, privacy protections and audit logs work in practice.
  • Whether any data providers or model developers begin using the proposed system.

Discussion spark: Would you trust a system that brings AI models to sensitive data, or should organisations keep that data out of external model training altogether?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.