MUMBAI: India’s AI ambitions are getting a safety net before they reach full speed. The government has approved 13 Responsible AI projects under the Safe & Trusted AI pillar of the IndiaAI Mission, stepping up efforts to make artificial intelligence safer, fairer and more transparent as the country accelerates its AI ecosystem.
The projects, spanning bias mitigation, privacy-preserving AI, explainability, machine unlearning, deepfake detection and AI risk assessment, were disclosed by Union Minister for Electronics and Information Technology Ashwini Vaishnaw in a written reply in the Rajya Sabha on 24 July.
The Safe & Trusted AI pillar is one of the seven pillars of the IndiaAI Mission, which was approved in March 2024 with an outlay of Rs 10,371 crore over five years to strengthen India’s artificial intelligence capabilities.
Among the approved initiatives is Saakshya, a multi-agent deepfake detection framework being developed jointly by IIT Jodhpur and IIT Madras. Other projects include AI Vishleshak, designed to detect audio and video manipulation, and IIT Kharagpur’s real-time voice deepfake detection system.
The programme also extends to responsible AI research across academia. NIT Raipur is developing bias mitigation algorithms for healthcare, while IIT Delhi, in collaboration with IIIT Delhi and IIT Dharwad, is building privacy-preserving machine learning models using federated learning. The Defence Institute of Advanced Technology (DIAT) is working on explainable and privacy-preserving AI for security applications, while IIT Jodhpur is developing machine unlearning techniques that allow generative AI models to forget sensitive or outdated information.
According to the Ministry of Electronics and Information Technology, the projects are intended to make AI systems fairer, more secure, transparent, privacy-preserving and trustworthy.
To strengthen oversight, the government has established the IndiaAI Safety Institute, which will supervise projects under the Safe & Trusted AI pillar and spearhead indigenous research on AI safety and governance. It has also issued India AI Governance Guidelines, proposing a risk-based framework to tackle algorithmic bias, misinformation, deepfakes and other societal challenges posed by AI.
The announcement comes as the broader IndiaAI Mission gathers momentum across infrastructure, research and talent development.
The government said it has identified support for 20 indigenous sovereign AI model proposals, including 12 large language models (LLMs) and eight small language models (SLMs). These include Sarvam AI’s 30-billion and 105-billion parameter models, Gnani.AI’s speech-to-speech model, BharatGen’s multilingual foundation models and Avataar AI’s video generation model.
On the infrastructure front, 15 compute service providers have been empanelled, while 237 AI projects have received support under the affordable compute programme, with 93 lakh GPU hours sanctioned.
The mission’s application development programme has also launched 12 national hackathons and innovation challenges, resulting in 62 AI prototypes and 20 AI solutions being developed.
India is simultaneously investing in AI talent. So far, 686 fellowships have been awarded across undergraduate, postgraduate and doctoral programmes in 178 institutions, while more than 26 lakh individuals have completed the YUVA AI for All programme.
The country’s AI infrastructure is also expanding beyond classrooms and research labs. The government has established 27 India Data and AI Labs, with work underway on another 188 labs, alongside plans to set up 58 Artificial Intelligence Centres of Excellence across states and Union Territories in partnership with industry and state governments.
As governments worldwide race to build more powerful AI systems, India is attempting to ensure safety keeps pace with innovation. By pairing investment in indigenous models and computing infrastructure with safeguards against bias, deepfakes and privacy risks, the IndiaAI Mission is signalling that the country’s AI ambitions are not just about building smarter machines, but also more trustworthy ones.
