India proposes DEPA framework for secure AI training

India's government has proposed a plan to regulate and secure information flow for AI development. The Ministry of Electronics and Information Technology named the scheme the Data Empowerment and Protection Architecture for AI Training. The motive is to create AI models trained on local data, as the government believes offshore model-makers are unlikely to address India's specific issues or ensure data owners retain control.

Professor Ajay Kumar Sood described the system as one where the model travels to the data. The scheme envisions Training Data Providers sharing anonymized and encrypted data with a Data Exchange Platform. Training Data Consumers then access this data to train their models within hardware-based Secure Data Environments. These environments keep data encrypted and isolated during processing, apply differential privacy techniques, and immutably log all activities for audit purposes.

MeitY envisions using the framework to unlock fragmented health data for diagnostics and drug discovery. It also aims to enable secure sharing of financial data to extend formal credit to underserved individuals and MSMEs. The framework states DEPA can overcome data interoperability challenges for personalized learning in the EdTech sector. It describes successful adoption as an essential catalyst for inclusive growth and responsible AI leadership, positioning the initiative as India's contribution to global science diplomacy.