As artificial intelligence reshapes industries globally, India is signaling a major policy shift. The government is now considering a formal legislative framework that could require pre-release vetting for powerful AI models.
From Advisories to Actual Law
India's approach to AI governance
is evolving. For the past few years, the Ministry of Electronics and Information Technology (MeitY) has relied on advisories under existing IT rules to guide platforms. These advisories urged companies to label AI-generated content, prevent the spread of misinformation, and tackle deepfakes. However, recent statements from top officials, including Union IT Minister Ashwini Vaishnaw, indicate a belief that the risks from AI have crossed a threshold where advisories are no longer enough. The government is now drafting a consultation paper, expected by November 2026, which will form the basis of a new, dedicated law for AI. This marks a significant move from a light-touch approach to building a formal regulatory structure.
What is 'Model Vetting'?
At the heart of the proposed legislation is the idea of 'vetting' or testing certain AI models before they are made available to the public. This would be a significant departure from the current model where developers can release systems freely. The plan, as reported, involves several potential layers of scrutiny. This could include pre-release evaluations to understand a model's capabilities, testing in controlled 'sandbox' environments to identify risky behaviors before they can cause harm in the real world, and a clearer allocation of responsibility for when an AI system causes damage. The focus appears to be on the most powerful 'frontier' models—those at the cutting edge of AI capabilities that could pose systemic risks, such as discovering software vulnerabilities or enabling sophisticated cyberattacks.
The Rationale: Safety and Trust
The government's primary motivation is to ensure AI safety and build public trust. Officials have explicitly pointed to the rising threat of high-quality deepfakes, financial fraud, and the potential for AI-driven misinformation to cause societal harm. There is also growing concern about 'agentic' AI systems—autonomous programs that can perform tasks with limited human supervision. Recent incidents where AI agents reportedly breached company networks without detection have added urgency to the regulatory push. By creating a framework that holds developers accountable, the government aims to establish clear rules for safety and prevent the misuse of powerful AI tools, ensuring a 'human-first' approach to the technology's deployment.
The Innovation vs. Regulation Dilemma
While the push for safety is understandable, the tech industry is watching with caution. The central debate is how to regulate AI without stifling the innovation that is crucial for India's ambition to become a global tech leader. A heavy-handed, bureaucratic approval process could slow down development, putting Indian startups at a disadvantage against global competitors. Industry experts and startup founders have voiced concerns that overly stringent guardrails could kill innovation, drawing parallels to how some startups moved out of Europe following its strict AI Act. Policymakers seem aware of this delicate balance, with officials noting the need to avoid placing a disproportionate compliance burden on smaller developers and research institutions. The forthcoming consultation paper will be a critical forum for the industry to provide feedback on how to craft a framework that is both safe and pro-innovation.
A Risk-Based Global Trend
India's proposed approach aligns with a global trend toward risk-based AI regulation. Rather than a onendize all law, many countries are looking to stratify rules based on potential harm. The European Union's AI Act, for example, imposes strict obligations on 'highrdsk' AI systems used in critical sectors like healthcare and law enforcement, while leaving lowrdsk applications largely unregulated. India appears to be considering a similar riskproportional model, where frontier models and highstakes applications would face mandatory audits and scrutiny, while less powerful tools retain flexibility. This strategy aims to focus regulatory resources where they are needed most, ensuring robust oversight for potentially dangerous AI without creating unnecessary hurdles for the vast majority of beneficial AI applications.
