From Policy Intent to Implementation: What Should India's AI Safety Institute Actually Do?

Abstract

India is at a defining moment in its AI trajectory. Artificial intelligence is already being explored and deployed across public service delivery, financial infrastructure, healthcare systems, and population-scale digital platforms. At the same time, India has set out an ambitious vision to become a global AI innovation leader through the IndiaAI Mission. Together, rapid deployment and national ambition create a difficult governance challenge. This paper argues that India's central AI governance question has changed. The question is no longer whether India needs an AI Safety Institute; the real question is how such an institute should work in practice. Drawing on global AI governance models, including the EU AI Act, the NIST framework, the UK and US AI Safety Institutes, and Singapore's AI Verify, the paper identifies what India can borrow and what it should avoid. The paper proposes a hub-and-spoke model for the IndiaAI Safety Institute: a central body that sets standards, supported by sectoral and regional cells that handle evaluation, implementation, and local expertise. It recommends five first-priority functions: risk-tiered classification of AI applications, mandatory safety audits for high-impact public-sector AI, Indian-language benchmarking, a national AI incident reporting system, and Safety-as-a-Service support for startups and smaller organizations.

Keywords