India weighs pre-release safety checks for frontier AI models, consultation paper expected November 2026

The regulatory trajectory in India is shifting toward requiring advanced AI system builders to demonstrate safety before launching their most powerful creations, potentially increasing compliance expenditures and slowing down product releases, which could affect India's standing in global AI competition.

The government is currently examining a distinct safety structure specifically for what are termed frontier models. This means developers might be required to show sufficient protective measures prior to releasing systems possessing the potential for severe harm. This represents a significant departure for a nation that previously favored soft guidance over strict regulations. This development was brought to light in a report from NDTV, which referenced sources familiar with the ongoing discussions. This initiative is connected to a consultation paper on AI governance that the government is preparing, and this document will ultimately shape India's upcoming AI legislation.

According to the research findings, the framework under consideration might incorporate multiple levels of examination. It is important to note that these measures are not yet finalized; they are under review, and the consultation period is specifically intended to gauge industry and public reactions to them. The concrete concerns prompting this dialogue involve the capacity of frontier models to facilitate cyberattacks and to generate synthetic material that appears extremely authentic.

Union Electronics and IT Minister Ashwini Vaishnaw has stated that the consultation paper will be made available in the coming weeks, soliciting input from civil society, citizens, industry, and other interested parties. The research findings suggest the paper could be available as early as November 2026, with the minister's announcement anticipated between October 23 and 28, 2026.

At least one major AI developer has reportedly already become involved in these discussions. Anthropic has allegedly held conversations with the Ministry of Electronics and Information Technology (MeitY) regarding safety testing and model access, and the firm has reportedly pushed for rules mandating transparency.

This proposal marks a notable change in India's previous stance. In November 2025, the government issued the India AI Governance Guidelines, a document that stressed voluntary and risk-based approaches. While that document recognized the growing risks associated with AI, it refrained from suggesting a standalone AI law. Less than a year later, the discourse has progressed toward formal legislation that imposes specific duties on the highest-capability systems.

Some foundational work is already in place for this regulatory shift. The IndiaAI Safety Institute is currently operational, and the research findings indicate that its supervisory function may be broadened as the regulatory discussions advance. Furthermore, security agencies have raised concerns; authorities such as CERT-In, India's computer emergency response team, have issued advisories and warnings about dangers linked to agentic AI—systems designed to perform actions on behalf of a user rather than merely providing answers.

For organizations developing frontier systems, the most immediate consequence pertains to compliance. Should pre-release evaluations and sandbox testing become mandatory requirements, developers would need to allocate both time and resources for these activities, which could consequently increase operational expenses and extend product deployment schedules. The extent of this required effort will hinge on the specific details, and how a frontier model is ultimately defined will be one of the most pivotal elements of the eventual statute.

From a competitive standpoint, the research findings observe that India's regulatory direction mirrors global trends in AI governance and takes cues from US-style methods for advanced AI. Several key points are now clearly identifiable to monitor: first, the release of the consultation paper and its precise definition of a frontier model; second, the extent of any expanded responsibilities assigned to the IndiaAI Safety Institute; and finally, the reaction from industry, which will indicate how much of the suggested structure will survive engagement with the targeted companies.