PocketEdge
← All news

Desk note · industry · Technology policy / artificial intelligence

MeitY published a handbook on sharing data safely for AI training

The Ministry of Electronics and Information Technology on 31 October 2026 released a Handbook on Data Empowerment and Protection Architecture for AI training at a Bengaluru roundtable, outlining how institutions can pool datasets for model building while keeping consent, privacy, and intellectual property controls in place.

31 Oct 2026 · 2 min read

Key takeaways

  • •MeitY launched the DEPA for AI Training handbook on 31 October 2026 with support from the iSPIRT Foundation at a Bengaluru roundtable chaired by Secretary S. Krishnan.
  • •The framework describes how organisations can share data for AI model training under contractual guardrails rather than open dumps of personal or proprietary records.
  • •India’s AI push depends on usable domestic datasets; the handbook is guidance, not a mandatory law, and adoption will turn on how banks, hospitals, and platforms design consent flows.

What's going on here?

On 31 October 2026, the Union Ministry of Electronics and Information Technology released the Handbook on Data Empowerment and Protection Architecture for AI training at a high-level roundtable in Bengaluru. Secretary S. Krishnan unveiled the document with support from the iSPIRT Foundation. The handbook explains how companies, universities, and government bodies can collaborate on datasets used to train artificial intelligence models while keeping privacy, intellectual property, and institutional control intact. It builds on India’s earlier DEPA work on consent-based data sharing, but targets the specific problem AI labs face: they need large, high-quality training sets without copying sensitive records into unsecured lakes.

What does this mean?

Model builders usually either scrape public web text or buy narrow commercial feeds. For Indian languages, payments, agriculture, or health, those shortcuts run out quickly and often break privacy rules. A DEPA-style architecture lets a hospital or bank remain the custodian of its records while granting a fintech or research lab access to derived features or anonymised slices under signed terms. That shifts the bottleneck from raw storage to governance design: who approves a request, how long access lasts, and what happens when a model is exported. The handbook does not by itself open any database; it gives technical and policy language for institutions that already hold data to decide whether and how to participate.

Why should I care?

If you run a regulated business sitting on customer or patient data, expect vendors to cite this handbook when pitching federated learning or synthetic data projects. Read the consent and audit clauses before you sign. If you invest in Indian AI or software services, ask whether revenue models assume free public data or paid, consented feeds. Watch which ministries reference DEPA in procurement rules, whether sector regulators such as RBI or IRDAI align their outsourcing norms, and whether state data centres pilot shared sandboxes under the architecture.

More desk notes

Disclaimer

For informational purposes only. Not investment advice. Verify with filings and a licensed adviser.