India’s Ministry of Electronics and Information Technology proposes a new architecture for AI training that keeps data encrypted and local.
What was released or claimed
The government describes the Data Empowerment and Protection Architecture for AI Training as a system where the model travels to the data.
Who it affects
Hospitals, banks, and government agencies would act as Training Data Providers, while AI developers serve as Training Data Consumers.
What is still unknown
It remains unclear how much work remains to implement the DEPA framework, despite existing data exchanges in India.
So what
The plan aims to ensure offshore model-makers address local issues while keeping data owners in control of their information.
What the paper shows
The proposed framework tests a model-centric approach to data access, reporting that hardware-based secure environments keep data encrypted during processing. It states that differential privacy techniques prevent models from learning individual-specific information. The material notes that all activities are immutably logged for audit purposes, ensuring transparency. The source admits it is unclear how much work remains to make the scheme a reality.
Why it matters
- Researchers must consider how encrypted, isolated data access changes model training pipelines and validation methods.
- The use of differential privacy in secure environments offers a concrete method for reducing individual data leakage in training.
- Immutable logging of secure environment activities provides a potential standard for auditing AI training processes.
- Local data control mechanisms may influence how global models are trained on region-specific datasets.
Topics: AI training, data privacy, India, DEPA, secure environments
Updated
Then / now / next
India operated several data exchanges without a specific framework for AI training data security.
India proposes DEPA for AI Training to keep data encrypted and local within secure environments.
Researchers should watch for implementation details and technical specifications of the secure data environments.
Story timeline
- Then India operated several data exchanges without a specific framework for AI training data security.
- 02 Nov 2026 India floats AI plan under which ‘The model travels to the data, rather than the data travelling to the model’
- Now India proposes DEPA for AI Training to keep data encrypted and local within secure environments.
- Next Researchers should watch for implementation details and technical specifications of the secure data environments.
For readers
Researchers
- Must adapt to encrypted, isolated data access for model training.
- Can study differential privacy implementations in secure environments.
Policy
- Offers a template for local data control in AI training.
- Addresses concerns about offshore model-makers ignoring local issues.
Developers
- Need to design models that travel to data rather than vice versa.
- Must integrate with secure data environments and immutable logging.
Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
How does DEPA for AI Training work?
It uses secure data environments where models access encrypted data without the data leaving the provider's control.
Who provides the data in this system?
Training Data Providers, such as hospitals and banks, share anonymized and encrypted data via a Data Exchange Platform.
Does the data leave the secure environment?
No, the data remains encrypted and isolated even during processing, with the model traveling to the data instead.
Connected stories
Reading receipt
- India’s MeitY proposes DEPA for AI Training to keep data local and encrypted.
- Secure environments isolate data while allowing model access for training.
- Differential privacy techniques aim to prevent individual data leakage.
Still open: How will the secure data environments technically enforce encryption during processing?
Still open: What specific differential privacy techniques will be systematically applied?
