New Delhi, Oct 7 The government on Tuesday announced the selection of five projects under the second round of its expression of interest for the 'Safe and Trusted AI' pillar under the programme launched by IndiaAI.
The selected projects will advance real-time deepfake detection, strengthen forensic analysis, address bias in AI models, and build robust evaluation tools for generative AI, ensuring that AI systems deployed in India are reliable, secure, and inclusive, according to an official statement.
A business division under the Ministry of Electronics and IT, IndiaAI, announced that the projects were selected for governmental support from over 400 proposals submitted by academic institutions, startups, research organisations, and civil society groups, after a review by a multi-stakeholder technical committee.
Collectively, these projects translate “Safe and Trusted AI” into practice, combining resilience testing and bias audits to support the responsible development and deployment of AI, the statement said.
The selected projects will focus on real-time voice deepfake detection, analysis for audio-visual and signature forgeries, evaluation of gender bias in agricultural large language models, and the development of penetration-testing tools for large language models and generative AI.
A multi-agent retrieval-augmented generation framework for deepfake detection and governance will be spearheaded by the Indian Institute of Technology Jodhpur in collaboration with IIT Madras. IIT Mandi and the Directorate of Forensic Services in Himachal Pradesh will develop AI Vishleshak, improving audio-visual deepfake detection and signature forgery detection, the ministry informed.
IndiaAI, a division of MeitY, serves as the implementation agency for the IndiaAI Mission. The mission focuses on democratising AI benefits, enhancing India's leadership in the field, promoting technological self-reliance, and ensuring ethical and responsible AI usage, the statement noted.
Disclaimer: This post has been auto-published from an agency feed without any modifications to the text and has not been reviewed by an editor