DEPA AI training is the centre of a new framework from India’s Ministry of Electronics and Information Technology (MeitY) that would let developers train artificial intelligence models on sensitive data without taking the underlying records away. The Data Empowerment and Protection Architecture for AI Training was outlined in a handbook launched on 31 October in Bengaluru, and it rests on one idea: the model travels to the data, rather than the data travelling to the model.
MeitY Secretary S Krishnan introduced the DEPA AI training framework at a roundtable in Bengaluru organised with support from the iSPIRT Foundation. Ajay Kumar Sood, Principal Scientific Adviser to the Government of India, took part in the launch and described DEPA in those terms. The handbook is marked as a draft for discussion and invites recommendations from the public. It does not replace the Information Technology Act, 2000, or the Digital Personal Data Protection Act, 2023, and it is not legally binding at present.
How DEPA AI training is meant to work
Under DEPA AI training, hospitals, banks and government agencies would act as Training Data Providers, sharing anonymised and encrypted data with a Data Exchange Platform. Model developers, called Training Data Consumers, would use the data held at those platforms to train their systems. The exchange would enforce each provider’s usage rules, so it would do more than list available datasets.
The platforms would run hardware based Secure Data Environments in which data remains encrypted and isolated even during processing. Developers would send training work into a controlled environment near the data, rather than receiving records for use on their own infrastructure. The intended output is a trained model, not a copy of the underlying dataset. The framework says differential privacy techniques are to be applied systematically to prevent a model from learning or revealing information specific to an individual, and that all activity inside the secure environments is to be logged in records that cannot be altered, for audit purposes.
MeitY’s motive, as described in coverage of the framework, is to create AI models trained on local data. The ministry’s view is that offshore model makers may lack the incentive to address India’s problems or to keep data owners in control of their information. The framework also sits within India’s digital public infrastructure approach, in which government backed shared tools are intended to stop large technology companies from dominating local markets. The Unified Payments Interface is the best known example of that approach.
Three sectors DEPA AI training aims to reach
Health is the first sector named in the framework. MeitY says DEPA AI training could unlock fragmented health data to support AI driven diagnostics, drug discovery and public health management. Hospitals hold large volumes of records, but those records are scattered across institutions and systems, which makes them hard for developers to use today.
Finance is the second sector. The framework says secure sharing of diverse financial data could extend formal credit to 400 million underserved individuals and micro, small and medium enterprises, and support new fintech products. A pilot project cited in coverage of the launch involved a bank that tested the system to identify customers eligible for credit cards, processing more than 2 million requests in a two week period.
Education is the third sector. The framework says DEPA AI training could overcome gaps between data systems to support personalised learning and portable skill credentials for India’s 250 million students, in an education technology sector valued at 29 billion dollars. The framework estimates that the approach could help unlock up to 500 billion dollars in AI driven economic value for India by 2030. That figure is the framework’s own estimate of potential, not a confirmed outcome.
What happens next for DEPA AI training
Consultations on putting DEPA AI training into practice are set to begin in the coming weeks, as stakeholders work towards a legal framework. Management of DEPA is expected to be overseen by Samyog, a new industry group supported by iSPIRT. The timeline for binding regulations is still uncertain, and the framework itself is described as a living resource that will change as the wider DEPA work develops.
DEPA builds on an earlier framework introduced in 2020 by NITI Aayog, which allowed citizens to share their data securely with consent. India already operates several data exchanges, but it is unclear how much work remains to turn the AI training architecture into an operating system. The framework also presents the effort as a contribution to global science diplomacy, citing the work of institutions such as the World Bank and the United Nations on a more inclusive global AI ecosystem. The DEPA AI training handbook asks for actionable recommendations from the public before any of its proposals become rules.