What Is This New AI Handbook?
The Ministry of Electronics and Information Technology (MeitY) has launched the 'Handbook on Data Empowerment and Protection Architecture (DEPA) for AI Training'. Unveiled in Bengaluru on October 31, 2026, this document is a techno-legal framework designed
to guide the responsible use of data in training AI models. It's not a new law, but a detailed guide for startups, researchers, and large enterprises. The goal is to address persistent concerns around privacy, intellectual property, and data control that often hinder access to high-quality datasets necessary for building robust AI. The framework was developed with support from the iSPIRT Foundation, a technology think tank.
Data Rights and Control at the Core
The handbook's central philosophy is enabling data sharing without data owners losing control. It proposes a novel approach where the 'model travels to the data, rather than the data travelling to the model'. This means AI developers can train their algorithms on sensitive datasets—like those held by banks or hospitals—within a secure, controlled environment without ever directly accessing or transferring the raw information. This method is designed to protect individual privacy and commercial confidentiality while still unlocking the value of large-scale data. The framework also mandates robust 'data provenance' mechanisms, which track the origin, permissions, and usage of data throughout the AI training lifecycle, ensuring transparency and accountability.
A Rulebook for Responsible Innovation
For AI developers and companies, the handbook provides a much-needed roadmap for compliance. It combines contractual permissions with privacy-enhancing technologies to create a trusted ecosystem for collaboration. Institutions that hold valuable data, referred to as 'Training Data Providers', can make their datasets available to 'Training Data Consumers' like startups and academic researchers. This exchange would happen through regulated intermediaries called 'Data Collabs'. The framework requires that all data sharing is purpose-bound, meaning a dataset used for training a healthcare model cannot be repurposed for financial services without new, explicit consent. This structure is intended to align with India's broader Digital Personal Data Protection (DPDP) Act of 2023.
Balancing Ambition with Guardrails
MeitY's initiative is a crucial part of India's larger national strategy on AI. MeitY Secretary S Krishnan stated that India's AI ambitions must be matched by the ability to use data responsibly. By providing clear, principle-based guidance rather than a rigid, top-down law, the government aims to foster innovation while building guardrails against potential misuse. The DEPA handbook is currently a draft for discussion, meaning the rules can change based on feedback from the industry and public before any official portal or platform goes live. This approach reflects a 'light-touch' regulatory philosophy that seeks to avoid stifling a fast-growing sector with heavy compliance burdens from the outset.
What Happens Next?
The release of the handbook marks the beginning of a new chapter in India's AI journey. While it's a significant step, the framework is currently a set of guidelines, not an operational system. The next phase will involve gathering public feedback, refining the principles, and eventually, building the technological and institutional infrastructure to bring the DEPA for AI Training to life. The success of this framework will depend on how effectively it can be implemented and whether it gains the trust of both data providers and AI developers. For now, the document serves as a clear statement of intent, positioning India as a thought leader aiming to shape a responsible and inclusive global AI ecosystem.
