Delhi wants critical departments to lean on India-built AI models, officials say
Senior officials say the central government wants departments in sectors such as defence and aerospace to use more AI built in India under the IndiaAI Mission, and may steer some away from non-Indian frontier models for critical applications. The details below are drawn from the primary source, with analysis labelled separately.
Senior government officials have said the Centre wants to increase the use of AI systems developed in India, especially in government-owned and critical sectors such as defence and aerospace, according to a report published on 10 November.
The push covers indigenous large language models, smaller sector-specific models and AI agents built under the IndiaAI Mission. Central and state departments are being asked to consider domestic options, and one official said some could be encouraged to avoid frontier models and agents from non-Indian companies for certain critical uses.
The government is also in talks with selected companies about whether existing models can be scaled up for crucial sectors. The IndiaAI Mission, with an outlay of ₹10,372 crore, has approved 20 models and chosen 12 startups to build them. Sarvam and BharatGen have released several models, and others named include Varya, EKA-1, Dhīr 2B and Dhīr 5B. Gnani, Gan AI, Fractal, Tech Mahindra and Shodh AI are also building.
Officials said more models are expected over the next 12 to 15 months, and that some domestic models have reached up to 80 per cent of the capability of leading frontier systems. The report quotes no named officials, so the details rest on unnamed sources.
Why it matters
This signals a procurement preference rather than a published rule, but it could create a protected early market for Indian model builders in sensitive domains. The 80 per cent capability claim shows why the policy is framed around critical use cases, not general-purpose work.
Departments will need evaluation methods that compare domestic and frontier models on the actual task rather than on headline benchmarks.
Indicops view
Public-sector buyers should expect sovereignty requirements to enter tenders and prepare model-agnostic architectures: abstraction layers, portable prompts and task-level evaluation sets let you swap an Indian model in for a foreign one without a rebuild. Enterprises serving government should plan for the same.
More in Government AI
Back to the newsroomPune's Froncort.AI wins $2.3 million from Google.org to take AI-guided TB screening to eight districts
Pune healthcare AI company Froncort.AI has received $2.3 million from Google.org to scale mAI-NETRA, an offline, local-language tool that directs frontline health workers to high-risk areas. A Pune pilot screened far fewer people per confirmed TB case. The details below are drawn from the primary source, with analysis labelled separately.
Andhra Pradesh clears Robotics and Physical AI Policy 2026-36 and land for a Reliance AI data centre
Andhra Pradesh's cabinet has cleared a ten-year Robotics and Physical AI Policy, allotted about 25 acres at Bhogapuram to Reliance for a gigascale AI data centre and approved a Rs 570 crore Epack unit. Incentive details are yet to be published.
MeitY circulates draft DEPA-for-AI training framework with pre-deployment checks for models
India's IT ministry has released a draft framework for training AI models on sensitive datasets inside secure environments, with models vetted before deployment. Press reports say participation is voluntary for now, while legal experts warn the checks could evolve into a licensing regime.