India weighs mandatory pre-release safety tests for frontier AI models

Crypto Briefing reports that India is considering a separate safety framework for frontier AI models, under which developers of the most capable systems might have to show adequate safeguards before release. The outlet says the plan surfaced in an NDTV report citing officials familiar with the discussions, and that it is tied to a government consultation paper on AI rules, now in preparation, which will feed into India's coming AI law.

The article lists layers of scrutiny that may be included: pre-release evaluations of what a model can do, testing in controlled environments so risky capabilities appear in a lab, and sandbox requirements allowing supervised trials. It stresses that nothing is final, since the measures are only being examined. The concerns behind the idea are that frontier models could enable cyberattacks and produce highly realistic synthetic content. Minister Ashwini Vaishnaw, who leads electronics and IT, has said the consultation paper will appear within weeks and seek input from industry, civil society and citizens; the article places that as early as November 2026.

The piece describes this as a shift. India's AI Governance Guidelines, issued in November 2025, relied on voluntary and risk-based measures and did not recommend a standalone AI law. The IndiaAI Safety Institute is already operating and its oversight role could grow. Anthropic has reportedly held talks with MeitY, the electronics and IT ministry, about model access and safety testing and has reportedly argued for mandatory transparency rules. CERT-In and other authorities have also warned about agentic AI.

Crypto Briefing says developers would face added costs and timelines if testing becomes mandatory, and that the definition of a frontier model, the Safety Institute's mandate and industry response are the points to watch.