India's DEPA Handbook Proposes Letting AI Models Train Inside Hospitals' and Banks' Data Vaults Instead of Copying the Data Out

India is trying to settle the training-data question with plumbing rather than a statute. The DEPA for AI Training handbook is aimed at data that never touched the open web, the records held by hospitals, banks and public agencies, and proposes that models be trained where that data sits instead of copying it out. That makes it closer to the country's payments and consent infrastructure than to the copyright fights over scraped text, and it arrives while the government is still preparing a consultation paper on a new AI law.

What MeitY released#

The Ministry of Electronics and Information Technology has launched a handbook on the Data Empowerment and Protection Architecture (DEPA) for AI training, an official release said on Saturday, according to PTI. IT Secretary S Krishnan unveiled it at a roundtable in Bengaluru, and it was developed with support from the iSPIRT Foundation, PTI reported.

The framework combines contractual permissions, controlled computing environments and privacy-enhancing safeguards, and adds data provenance so that the origin, permissions and use of data can be traced through training, the release said, as quoted by PTI and ETV Bharat. Principal Scientific Adviser Ajay Kumar Sood said that under DEPA the model travels to the data rather than the data travelling to the model, ETV Bharat reported.

How it would work#

Business Standard, which read the underlying report, describes four technical pillars: a Data Exchange Platform, Electronic Contracts, a Secure Data Environment and Differential Privacy. Data providers such as hospitals, banks and public agencies would register datasets, and startups, researchers and enterprises would request access for a specific use case, according to Business Standard's account. Personal and sensitive data would not be shared in raw form, and trained models would face bias, security and explainability checks before deployment, the report says.

The report distinguishes DEPA Training from DEPA Inference, which works through the Account Aggregator model in finance, Business Standard reported. Its roadmap starts with pilots in finance and healthcare, with regulatory sandboxes in parallel, and the report claims the framework could unlock $500 billion in AI-driven value for India's GDP by 2030.

What the handbook itself says#

The handbook posted on MeitY's website is titled "Draft for Discussion" on its first page. It does not propose a new law. It says DEPA for Training would align with existing ones, naming the IT Act, 2000 and the Digital Personal Data Protection Act, 2023, and it calls for an oversight body to set privacy, ethics and security standards, register participants, oversee data-sharing procedures and resolve disputes. It does not say who would run that body. As evidence the approach works, it cites a Penn Medicine and Intel project in which 71 institutions used federated learning to train a brain-tumour model without sharing raw patient data.

The plumbing is older than the handbook. iSPIRT's open-source reference implementation, depa-training, has been public on GitHub since September 2023 under a CC0 public-domain dedication.

That puts the open question in governance, not technology. Whether hospitals and banks register data will depend on who sits on the oversight body and how disputes are settled, and the draft leaves both open. India's regulators have so far moved sector by sector, as SEBI did when it said it would issue AI guidelines for capital markets, and the handbook's finance and healthcare pilots would fall under those same sector regulators.

iSPIRT co-founder Sharad Sharma called the launch a more "significant moment since February 2016, when the Unified Payment Interface (UPI) was announced," ETV Bharat reported.

Telecom operators separately face TRAI's draft responsible-AI rules, which would govern how they use AI rather than how models are trained.

Limits#

• The government's press release was not available to read; launch details come from PTI, ETV Bharat, Business Standard and Hindustan Times. The handbook's contents are from the PDF on MeitY's website.
• Business Standard says Krishnan unveiled the handbook on Friday, October 30, and calls it a draft; PTI and ETV Bharat describe a launch on Saturday. The handbook is labelled a draft for discussion but gives no comment process or deadline.
• The $500 billion figure is the report's own estimate, not an independent forecast.

Sources#

• MeitY: Data Empowerment and Protection Architecture for AI Training, Draft for Discussion (October 2026)
• iSPIRT: depa-training reference implementation (GitHub)
• Hindustan Times: Govt rolls out data-sharing framework to boost AI training while protecting privacy (October 31, 2026)
• Business Standard (PTI): MeitY launches DEPA handbook to enable secure data sharing for AI training (October 31, 2026)
• ETV Bharat: Govt launches handbook on Data Empowerment and Protection Architecture for AI training (October 31, 2026)
• Business Standard: India proposes framework to unlock data for AI training, protect privacy (October 30, 2026)