India floats AI plan under which ‘The model travels to the data, rather than the data travelling to the model’
India has proposed a framework for training artificial intelligence models on sensitive data without requiring organisations to hand over their raw records.
By The Register
India has proposed a framework for training artificial intelligence models on sensitive data without requiring organisations to hand over their raw records.
The plan, called the Data Empowerment and Protection Architecture for AI Training, has been developed by India’s Ministry of Electronics and Information Technology.
Professor Ajay Kumar Sood, Principal Scientific Adviser to the Government of India, described the approach as one where “the model travels to the data, rather than the data travelling to the model”.
Under the proposed system, organisations including hospitals, banks and government agencies could act as Training Data Providers. They would share anonymised and encrypted information through Data Exchange Platforms.
AI developers and other approved users, known as Training Data Consumers, would send their models or training jobs into secure environments operated by those platforms. The data would remain inside the environment rather than being copied to the developer.
The framework says the secure environments would use hardware-based protections so information remains encrypted and isolated during processing. Data providers would also set rules governing how their information could be used.
MeitY’s framework proposes the use of differential privacy techniques to reduce the risk of models learning or revealing information about individuals. Activity inside the secure environments would be permanently logged to support audits and provide a record of how the data was handled.
The proposed system is intended to support the development of AI models using local data while keeping greater control with the organisations that hold it. India’s government believes overseas model-makers may not be sufficiently motivated to address the country’s specific needs or protect local data interests.
Possible applications include combining fragmented health records to support AI-assisted diagnosis, drug discovery and public health work. The framework also identifies financial services as an area where secure data-sharing could help develop new AI tools.
India already operates several data exchanges, but the extent of work still needed to put the proposed DEPA system fully into operation remains unclear.
The framework was launched in Bengaluru on 31 October 2026 and is presented as a proposed framework rather than binding legislation.