India bets on ‘closed rooms’ to train AI on sensitive data without revealing it | Today’s news

New Delhi/Bengaluru: India on Friday unveiled a system that allows companies and researchers to train artificial intelligence (AI) algorithms on sensitive data held by hospitals, banks and government agencies without the information leaving or being exposed to those institutions.

The Data Empowerment and Protection Architecture (DEPA) framework for AI training was launched by the Ministry of Electronics and Information Technology (MeitY) and developed primarily by iSPIRT, a volunteer-led technology think tank that contributed to India’s Unified Payments Interface (UPI) and India Stack, a set of digital secure public infrastructure systems enabling services such as digital data payments and digital identity (Dahacker).

The central idea of ​​Depa is to bring AI models to data, not to move data to researchers, a senior government official said at the unveiling of the framework in Bengaluru on Friday.

Read also | Trai proposes mandatory AI policy, incident reporting for telcos

The depot itself is not new. In 2020, NITI Aayog, the government’s top policy think tank, introduced it as a framework that allows citizens to securely share their data with their consent. A new framework, developed by iSPIRT, extends this approach to train artificial intelligence models on sensitive data belonging to millions of people without requiring the data to leave the institutions that hold it.

Consultation on Depa for AI will begin in the coming weeks before it is put into legal framework, a process that could take years, stakeholders have confirmed. The depo comes as India prepares to implement its first data protection law, the Data Protection Act 2023 (DPDP), in the coming months. Meanwhile, several lawsuits involving AI platforms and their use of proprietary data remain litigated.

The official mentioned above called the approach a “democratic third way” between the US, where companies buy data in trade deals – and China, where the state pools it.

Read also | AI Security Expert Nate Soares Stopped Saving for Retirement – Here’s Why

Industry stakeholders have identified the framework as potentially essential for effectively training domain-specific AI models, applications, and products.

“While the framework could potentially be vital, the real use cases could be for startups like Sarvam and others looking to create domain-specific AI for healthcare, education, critical communications and more. But the critical aspect would be to build a trusted framework beforehand, as well as validate the quality of large tranches of sensitive data – both of which are difficult to establish,” said Jibu Elias, Global Internet Foundation head, responsible computing in India.

Within the Depo, data remains with its owner, while in a “locked room”, a secure computing environment, only what is permitted by an authorized digital contract runs. No one can access the data inside, not even the company running the system, as the opening presentation states. The developer submits an AI model and receives only the trained model or, in cases such as a loan application, the decision itself.

Read also | As U.S. airlines deploy AI pricing tools, startups are helping travelers cash in on lower fares

On Friday, the first pilots of the framework were demonstrated. In one such attempt, the bank, which has been testing the approach since March 2026, identified which fintech customers qualified for its credit cards before asking anyone for approval, then only asked for permission from those who had a genuine offer. The pilot processed more than 2 million requests in two weeks. In another pilot project, a hospital and a healthcare startup are training a tumor detection model on MRI scans that the startup never sees.

The event was attended by Ajay K Sood, Chief Scientific Advisor to the Government, Secretary Meity S. Krishnan, iSpirt Co-Founder Sharad Sharma and representatives from HDFC Bank, Sarvam and Emergent AI among others.

Stakeholders said the framework could be key because hospitals and banks rarely share data. They earn little from it and face fines up to ₹250 crore under the Indian Data Protection Act if there is a leak.

India’s existing consent-based data-sharing framework “solves the individual consent problem brilliantly, but is unable to solve the collective problem,” a bank executive said, citing the multi-bank fraud as an example.

For the time being, the day-to-day management of Depa is expected to be handled by Samyog, a new industrial group “powered by iSpirit”. The website says it will register participants, provide standard contracts and resolve disputes. Its board includes Sharma, and several leaders helped write the manual. The site does not say that Samyog is appointed by the government.

iSpirt executives said data owners such as hospitals and banks will be paid each time their data is used. However, the manual only states that payment methods “will be investigated.” It also does not specify who would be liable in the event of a data leak from a locked room.

The framework is not legally binding, and officials have not said if or when it might be. The way forward includes pilot projects in finance and healthcare and regulatory sandboxes where companies test models under supervision. Binding rules are likely to come from sectoral regulators rather than through a new law, as was the case with India’s bill collection system, which took five years to move from RBI rulemaking to launch.

Much also depends on the Privacy Act of 2023, the main rules of which will come into effect in May 2027. India also continues to explore specialized AI legislation while debating whether AI platforms need a whole new definition of how they are protected under the law, Mint reported on Wednesday.