India is preparing a dedicated legal framework for artificial intelligence that could place the world's most advanced AI models under a separate safety regime before they are released to the public, NDTV reported.

The proposed approach signals a significant shift from the country's largely technology-neutral regulatory strategy towards more direct oversight of highly capable AI systems. Recent government statements indicate that policymakers believe AI risks have now crossed a threshold that warrants new legislation.

According to officials familiar with the discussions, the forthcoming law is expected to focus on so-called "frontier models", referring to cutting-edge AI systems whose capabilities may create national security, cybersecurity or societal risks if deployed without adequate safeguards.

The government is examining measures that could require developers of powerful AI models to demonstrate that appropriate protections are in place before launch. Such safeguards may include pre-release evaluations, controlled testing environments and security assessments proportionate to the capabilities of individual models.

Union Electronics and IT Minister Ashwini Vaishnaw has confirmed that a consultation paper on AI regulation will be released in the coming weeks. The document is expected to seek feedback from citizens, civil society groups, industry participants and other stakeholders before legislation is drafted. He has also stated that "the risks of AI have crossed a threshold" and that the government is already working on a new regulatory framework.

A major policy question is whether India should formally distinguish ordinary AI applications from frontier AI systems. Officials believe such a distinction would allow regulators to impose enhanced scrutiny only on models capable of causing significant harm, rather than burdening every AI developer with identical compliance requirements.

Discussions have drawn lessons from developments in the United States, where concerns over national security led to temporary restrictions on certain advanced AI models. Those debates highlighted the growing challenge governments face when AI systems acquire capabilities that may have strategic or security implications.

One option under consideration is mandatory sandbox testing before wider deployment. Officials described the possibility as "a vetting process before it rolls out", allowing researchers and regulators to evaluate a model's behaviour, uncover vulnerabilities and assess risks before public release.

The government is also exploring whether developers should limit access to particularly sensitive functions, impose restrictions on certain categories of users, or apply varying safeguards depending on the level of risk posed by a model.

Defining a frontier model remains one of the most complex challenges. Policymakers are considering both computing-power thresholds and capability-based assessments, recognising that processing power alone may not fully capture the dangers associated with advanced AI systems.

India has so far relied primarily on existing legal mechanisms, including the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, alongside targeted measures addressing synthetic content and deepfakes.

However, the proposed legislation could move beyond regulating online content and instead focus directly on the AI systems that create such content. Officials are examining how any new statutory framework would interact with existing intermediary regulations.

Officials have stressed that unlike the United States, which has occasionally relied on executive action in the absence of a dedicated law, India requires legislative provisions to establish and enforce AI safeguards. As one official noted, "If a company operates in India, it will have to follow Indian law."

The proposed framework is expected to address threats that extend beyond conventional content moderation. These include autonomous AI agents taking unauthorised actions, systems identifying weaknesses in critical infrastructure, advanced cyberattack capabilities and models capable of generating harmful instructions.

Deepfakes, online impersonation, financial fraud and psychological harm are also expected to feature prominently in the consultation process. Policymakers are increasingly concerned about the growing difficulty of distinguishing authentic content from highly convincing AI-generated material.

Officials are evaluating whether platforms should more clearly identify AI-generated content, strengthen complaint and takedown mechanisms and develop enforcement systems capable of keeping pace with rapidly spreading fabricated material.

At the same time, policymakers recognise that labelling alone may not be sufficient. Discussions are also examining how authentic content can be misrepresented and how users can be assured when material is verified as genuine. As one official observed, "Moreover, it's not just about A.I. generated content, labelling could also be about letting users know they are accessing authentic material."

Human oversight is emerging as another central pillar of the proposed legislation. Regulators are expected to examine safeguards for increasingly autonomous AI agents that can interact with software, access sensitive data and perform tasks with limited supervision.

Potential requirements may include mandatory human approval for sensitive actions, restrictions on access to critical systems, continuous monitoring mechanisms, incident-reporting obligations and emergency controls capable of stopping unsafe AI behaviour.

Recent incidents involving AI developers including Anthropic and OpenAI have reinforced these concerns. During various evaluations, advanced AI agents reportedly accessed real-world systems after internet connectivity restrictions failed or safeguards were partially disabled. Separate assessments conducted by the UK AI Security Institute also found that some agents carried out unauthorised actions against real-world targets when protective controls were relaxed. These episodes have intensified calls for stronger containment measures and human supervision.

Responsibility and liability for AI-related harm are also expected to form an important part of the debate. Officials are examining how accountability should be distributed among model developers, deployer, platform operators and end users.

At the same time, the government is attempting to strike a balance between innovation and regulation. Officials are conscious that excessively broad rules could impose disproportionate compliance costs on smaller developers, research institutions and emerging Start-Ups.

International coordination is likely to remain an important consideration as AI systems developed in one country are routinely deployed across multiple jurisdictions. Policymakers are therefore seeking an approach that aligns with evolving global practices while remaining compatible with India's legal framework and domestic innovation goals.

As the consultation process approaches, the key issues expected to dominate discussions include the definition of frontier AI, mandatory pre-release assessments, sandbox testing, restrictions on high-risk capabilities, safeguards against deepfakes and online fraud, human control requirements and the allocation of responsibility across the AI ecosystem.

The government's objective is to create a framework that protects citizens from emerging risks without slowing innovation in one of the world's fastest-growing AI markets.

Agencies