India's Ministry of Electronics and Information Technology launched a handbook for its Data Empowerment and Protection Architecture for AI training. It was unveiled at a roundtable in Bengaluru by Secretary S. Krishnan. The framework aims to unlock data for AI model training while protecting privacy.
How the framework works
The framework combines confidential clean rooms, differential privacy, electronic contracts, and zero-trust security. Training data providers would register encrypted, anonymized datasets on a data exchange platform. Consumers could request datasets, and electronic contracts would check their requests against data-sharing policies.
Key participants
The framework defines training data providers and training data consumers. Providers can include banks, hospitals, and government agencies. Consumers include startups, enterprises, researchers, and individuals.
Model training and oversight
The framework calls for models to train inside secure data environments without exposing raw data. Trained models would need to pass independent checks for bias, fairness, security, safety, and explainability. Approved models would receive clearance for pilot deployment and scaling.
MeitY developed the guidelines with iSPIRT Foundation. The roundtable included representatives from government, industry, technology, academia, research, and media. The ministry said it is committed to taking the framework forward to advance AI innovation in India.