MeitY Unveils DEPA Handbook for Trusted Data Sharing in AI Training

MeitY Introduces Privacy-First Framework for AI Development

The Ministry of Electronics and Information Technology (MeitY) has launched a comprehensive handbook on Data Empowerment and Protection Architecture (DEPA) for AI training, marking a significant step in India's approach to responsible and secure data usage1

1

. IT Secretary S Krishnan unveiled the framework at a roundtable in Bengaluru, developed in collaboration with the iSPIRT Foundation, emphasizing that "India's AI ambitions must be matched by the ability to use data responsibly and securely"1

1

. This consent-based framework aims to unlock data for training AI models while addressing critical concerns around privacy, confidentiality, and intellectual property that often limit data sharing2

2

.

Techno-Legal Architecture Enables Secure Collaboration

The Data Empowerment and Protection Architecture establishes a techno-legal framework that allows institutions to collaborate on data for AI training while maintaining control over their datasets. The framework combines contractual permissions, controlled computing environments, and privacy-enhancing safeguards to enable trusted data sharing for AI1

1

. Principal Scientific Adviser Ajay Kumar Sood explained a fundamental shift in approach: under DEPA, the model travels to the data rather than the reverse, enabling multi-party and cross-border collaboration1

1

. This architecture builds on India's existing DEPA framework, which already supports consent-based data sharing through initiatives such as Account Aggregator2

2

.

Technical Pillars Support Privacy-Preserving Data Sharing

The framework encompasses four critical technical pillars designed to protect data while enabling AI training. Confidential Clean Rooms create secure computing environments where organizations can collaborate on data processing without exposing underlying datasets to one another2

2

. Differential Privacy techniques ensure that individual-level information cannot be identified or extracted from aggregate datasets2

2

. Electronic contracts define machine-readable agreements that enforce data-sharing conditions, including permitted uses and security requirements2

2

. Zero-Trust Security authenticates and authorizes every access request, with encryption, continuous monitoring, and auditing to protect data throughout the process2

2

. The framework also incorporates data provenance, enabling the origin, permissions, and use of data to be traced throughout the training process, supporting accountability and transparency1

1

.

Stakeholder Ecosystem Facilitates Data Access

DEPA defines three key stakeholder categories to enable privacy-preserving data sharing. Training Data Providers (TDPs) are organizations like banks, hospitals, and government agencies that make their data accessible through the framework2

2

. Training Data Consumers (TDCs) include startups, enterprises, researchers, and individuals seeking access to datasets to train AI models2

2

. Data Collectives or Data Collabs serve as intermediaries that aggregate, standardize, and organize data from various sources while defining data-sharing policies and governance rules for participants2

2

. This structure helps startups, businesses, and academic institutions access large-scale, India-specific datasets to develop AI models while protecting citizens' privacy and ensuring compliance with the DPDP Act2

2

.

Workflow Ensures Compliance and Security

Under the DEPA framework, Training Data Providers first register their datasets on a data exchange platform after anonymizing and encrypting them, with datasets tagged with metadata and cryptographic identifiers for traceability2

2

. When Training Data Consumers request a dataset for specific use cases, an electronic contract system checks their requests against data-sharing policies and governance rules before granting access2

2

. Once access is granted, the model is trained within a secure data environment using differential privacy methods, ensuring that raw data remains inaccessible to consumers2

2

. Trained models must undergo independent bias checks for fairness, security, safety, and explainability, with only approved models receiving clearance for pilot deployment and scaling2

2

.

Long-Term Oversight and India's AI Governance Push

The framework ensures continuous monitoring even after deployment, with Training Data Consumers subject to periodic reviews and traceability of the model's outputs back to the data and contracts used in its training2

2

. This launch aligns with India's broader focus on AI governance, following TRAI's recent draft guidelines on responsible AI use in telecom networks and service delivery2

2

. Union IT minister Ashwini Vaishnaw has indicated that the centre will release a consultation paper for regulation on AI safety and security within a month2

2

. Principal Scientific Adviser Ajay Kumar Sood described DEPA as India's contribution to global science diplomacy, suggesting the framework could serve as a model for international collaboration on AI training1

1

. The roundtable drew representatives from government, industry, academia, and research stakeholders, indicating broad interest in implementing this approach to responsible and secure data usage for AI development1

1

.