MeitY secretary warns proprietary AI may move Indian data abroad, backs model mix
Representative image: cookieone / Pixabay
MeitY Secretary S Krishnan said on 8 October 2026 that proprietary AI models risk moving Indian data abroad and called for a mix including open-weight models.
What Happened
India's top electronics and IT official has warned that relying on proprietary artificial intelligence models carries a risk of Indian data leaving the country and of that data being used to train foreign systems. S Krishnan, Secretary of the Ministry of Electronics and Information Technology (MeitY), made the remarks on Thursday, 8 October 2026, in New Delhi, on the sidelines of the India launch of the World Bank Group's World Development Report 2026, an event organised with the IndiaAI Mission. He said the government's answer is not a single type of model but a combination, including open-source and open-weight models that can run without data being sent abroad.
Key Facts
- Speaker: S Krishnan, Secretary, Ministry of Electronics and Information Technology.
- Date and place: 8 October 2026, New Delhi, at the India launch of the World Bank's World Development Report 2026.
- Concern raised: proprietary models may move Indian data outside the country and may learn from the information users feed into them.
- Proposed answer: a mix of models, with open-source and open-weight models that can be run without sending data out of India; he said the government is working on this.
- On data localisation for government-empanelled AI firms, he said it would depend on the nature of the data, pointing to the government's data classification and its matching rules for data centres and cloud services.
- He said India is building the full AI stack, from compute and data to models and applications, to protect its strategic autonomy while staying open economically.
- The IndiaAI Mission has an outlay of about Rs 10,372 crore over five years.
Why It Matters
Most of the AI tools that Indian offices, banks, hospitals and government departments use today are proprietary models built and hosted by foreign companies. When a user types a question or uploads a document, that material is often processed on servers outside India. Krishnan's comments put on record that the government sees two separate risks in this: the physical transfer of data across borders, and the possibility that the model provider learns from what Indian users share.
His proposed remedy is pragmatic rather than a ban. Open-weight models are those whose trained parameters are published, so an organisation can download them and run them on its own machines or on a cloud located in India. That keeps sensitive information inside the country while still giving access to capable AI. Krishnan's phrase, "So there has to be a judicious mix," signals that proprietary tools will remain in use where the data involved is not sensitive.
The remark on data classification is important for companies that sell AI services to the government. India already sorts government data into categories with different storage rules for data centres and cloud providers. Krishnan's answer suggests that AI vendors will be judged by the same yardstick: the more sensitive the data, the stronger the case for keeping processing inside India. This links the AI debate to rules that already exist rather than to a fresh, sweeping law.
The comments came in a busy week for AI policy. At the same event, Electronics and IT Minister Ashwini Vaishnaw said a consultation paper on AI regulation would be released within about a month, and that 10,000 more graphics processors would be added under the IndiaAI Mission.
| Model type | Where it runs | Main data concern |
|---|---|---|
| Proprietary, hosted abroad | Provider's servers, often outside India | Data may leave India and be used for training |
| Open-weight, self-hosted | User's own or India-based servers | Data can stay inside India |
| Domestic models | Indian infrastructure | Depends on how the developer stores data |
Impact
Short-term: Government departments and public sector bodies are likely to look harder at where their AI tools process data, and vendors may be asked to explain data handling in more detail.
Long-term: Demand could grow for India-hosted open-weight models and for Indian-built models, supporting the IndiaAI Mission's compute and model programmes and local cloud providers.
Who is affected: Ministries and state departments buying AI services, Indian AI start-ups building on open-weight models, global AI companies selling to Indian institutions, and citizens whose records pass through public systems.
Key Takeaway
MeitY wants India to use a mix of proprietary and India-hosted open models, with the sensitivity of the data deciding where AI processing may happen.
Questions and Answers
What exactly did the MeitY Secretary warn about?
S Krishnan said on 8 October 2026 that proprietary AI models can carry Indian data outside the country and may learn from the information users give them.
Is the government banning foreign AI tools?
No. Krishnan called for a judicious mix of models, adding open-source and open-weight models that can run inside India alongside proprietary ones.
What is an open-weight model?
It is an AI model whose trained parameters are released publicly, so an organisation can run it on its own or India-based servers instead of sending data to the developer.
Will AI firms working for the government have to keep data in India?
Krishnan said it depends on the nature of the data, under the government's existing data classification and the matching rules for data centres and cloud services.
Disclaimer: Prepared by the Peepals newsroom from publicly available sources with AI assistance. Information is accurate to the best of our knowledge at the time of publication and may change. Images may be representative. Not professional advice. Report an error via our contact page.
Sourced and fact-checked by the Peepals Global Editorial Team
Gujarat to use student-built tools in policing after e-Rakshak cyber hackathon
Autonomous freight rail start-up Parallel Systems raises $100 million Series C
Reported, fact-checked and published by the Peepals Global Editorial Team.
