The days of indiscriminately scraping data to train AI may be ending. India has unveiled a new framework, the DEPA for AI Training, which aims to create a structured, secure, and consent-driven approach to data collaboration for innovators.
A New Blueprint for AI Data
Artificial
intelligence runs on data, but the most valuable datasets—containing financial, health, and personal information—are often locked away due to privacy and security risks. To solve this, India's Ministry of Electronics and Information Technology (MeitY) unveiled the Data Empowerment and Protection Architecture (DEPA) for AI Training on October 31, 2026. This isn't just another policy paper; it's a detailed technical and legal blueprint designed to change how AI models are built. The core principle, as described by top government officials, is that 'the model travels to the data, not the data to the model'. This means AI developers can use sensitive information without ever seeing or copying the raw data, a fundamental shift from today's common practices.
How 'The Model Travels to Data' Works
The DEPA framework proposes a new ecosystem with three main roles: Training Data Providers (like hospitals or banks), Training Data Consumers (startups and researchers), and sector-specific 'Data Collabs' that aggregate data. The system is built on four technical pillars. First, a Data Exchange Platform allows data providers to register their datasets. Second, machine-readable Electronic Contracts automatically enforce the rules of how data can be used. Third, all training occurs inside a Secure Data Environment (SDE), a kind of digital 'clean room' where the data remains encrypted and isolated. Finally, the framework employs Differential Privacy, a technique that adds statistical noise to prevent the re-identification of individuals, ensuring privacy is mathematically enforced. An AI developer sends their model into this secure environment to train on the data, but can only extract the improved model, not the underlying information.
Rethinking Consent for the AI Era
The framework makes a critical distinction between using AI for 'inference'—making a decision about one person, which requires explicit consent—and 'training' an AI model on bulk data. The DEPA handbook notes that seeking individual consent for training on millions of records is often impractical. Instead, it builds a system of trusted, auditable governance. Consent is managed at an institutional level, where data providers act as stewards. The combination of secure technology, contractual obligations, and privacy-enhancing techniques is designed to protect individuals' rights without halting innovation. This approach moves beyond simple anonymization, which has proven vulnerable, to a more robust model of 'compliance by design' where privacy isn't just a policy, but a core feature of the architecture itself.
Unlocking India-Specific Innovation
The strategic goal of DEPA is to fuel a homegrown AI ecosystem that can solve uniquely Indian challenges. The government estimates the framework could unlock $500 billion in AIndriven economic value by 2030. By providing secure access to high quality local data, DEPA aims to help startups and researchers build models for critical sectors. For example, models trained on Indian financial data could help extend formal credit to millions of underserved individuals and small businesses. In healthcare, AI could be trained on diverse clinical data to improve diagnostics for India's population. This initiative is seen as essential for creating AI that is not only powerful but also relevant and contextualised for the Indian market, reducing reliance on models trained on data from other parts of the world.
A 'Digital Public Good' for the World?
DEPA for AI Training is the latest chapter in India's Digital Public Infrastructure (DPI) playbook, following successes like the Unified Payments Interface (UPI) and Aadhaar. It's an assertion of 'digital sovereignty'—a belief that nations should control their own data destiny. By creating a sophisticated technical and legal framework, India is building an alternative to the dominant models of either statecontrolled data or corporate data extraction. The ambition extends beyond national borders. India has a history of offering its DPI models to other countries, and DEPA could become a template for developing nations that want to harness AI without surrendering their citizens' data to a few global tech giants. It represents a tangible blueprint for balancing innovation with privacy on a national scale.
