India weighs pre-release safety checks for frontier AI models

A forthcoming consultation paper could split India's AI rules into two tracks, with the most powerful models facing tests before launch India may soon ask the builders of its most powerful AI systems to prove those systems are safe before they ship. The government is examining a separate safety framework for so-called frontier models. Developers could have to demonstrate adequate safeguards before releasing systems capable of serious harm. That would mark a real change for a country that has so far preferred gentle nudges to hard rules. The plan surfaced in a report by NDTV, which cited officials familiar with the discussions. It is tied to a consultation paper on AI regulation that the government is preparing. That paper will shape India’s forthcoming AI law. Two lanes for AI The framework under consideration may include several layers of scrutiny, according to the research findings: Pre-release evaluations to assess what a model can do before it reaches the public. Controlled-environment testing, so risky capabilities surface in a lab rather than in the wild. Sandbox requirements, which would let developers trial models under supervision. None of this is final. The measures are b...