India's AI ambitions are massive, but they run on data—a resource fraught with privacy risks. A new handbook from the Ministry of Electronics and IT proposes a new way forward, aiming to balance innovation with individual rights.
What is the DEPA Handbook?
The Ministry of Electronics
and Information Technology (MeitY), with support from the iSPIRT Foundation, recently launched a handbook for what it calls the Data Empowerment and Protection Architecture (DEPA). This isn't a new law, but a detailed guidebook—a 'techno-legal framework'—for how institutions like hospitals, banks, and universities can share data with researchers and companies to train artificial intelligence models. The goal is to solve a major problem: AI developers need vast amounts of high-quality data, but data holders are often reluctant to share it due to concerns about privacy, intellectual property, and losing control. This handbook, currently a draft for discussion, lays out a set of rules and technical systems to make this sharing possible and safe.
The Model Travels to the Data
The core innovation proposed in the DEPA handbook is a fundamental shift in how data is accessed. Instead of data providers sending their sensitive datasets to AI developers, the framework is built on a principle described by Principal Scientific Adviser Ajay Kumar Sood as 'the model travels to the data'. In practice, this means an AI algorithm from a 'Data Consumer' (like a startup or researcher) is sent into a secure, controlled digital environment managed by the 'Data Provider' (like a hospital). The model is trained on the data within this sealed-off 'confidential clean room' without the raw data ever leaving the provider's control. This approach is designed to enable multi-party collaboration, even across borders, without centralizing and exposing sensitive information.
What Are the Key Guardrails?
The framework isn't based on trust alone; it's built on a combination of three key enforcement mechanisms. First, there are legally binding digital contracts that strictly define the purpose for which the data can be used. An AI model being trained for cancer detection, for example, cannot have its access repurposed for another task. Second is the use of controlled computing environments, or 'clean rooms', which technically prevent the data from being copied or moved. Third, the framework incorporates privacy-enhancing technologies like differential privacy, which adds statistical 'noise' to data to ensure no single individual can be identified from the dataset. This layered approach creates the 'guardrails' to protect both the data provider and the individuals whose information is in the dataset.
Placing Data Rights and Consent First
Central to the entire architecture is its alignment with India's Digital Personal Data Protection (DPDP) Act. The DEPA framework is built to operate within the consent-based regime of the DPDP Act. It introduces the concept of 'data provenance', which creates a clear, auditable trail for every piece of data. This means it's possible to trace where data came from, what permissions were granted for its use, and how it was applied throughout the training process. For individuals, this means their data is used based on specific, purpose-bound consent, which is managed through a machine-readable 'consent artefact'. This ensures transparency and accountability, making it a significant step beyond the often-vague privacy policies of the past.
The Road Ahead for India's AI Ambitions
While the handbook is a landmark proposal, it is currently a guide, not a mandatory regulation. The next steps will involve refining this draft with stakeholders from government, industry, and academia who were present at its launch. IT Secretary S Krishnan has stated that MeitY is committed to taking the framework forward, emphasizing that India's AI goals must be matched by the ability to use data responsibly and securely. If successfully implemented, the DEPA framework for AI training could become a cornerstone of India's Digital Public Infrastructure, similar to the role UPI played in payments. It offers a pathway to unlock valuable datasets trapped in institutional silos, potentially fueling a new wave of homegrown AI innovation while building a foundation of public trust.
