Govt to issue consultation paper on a new AI law in ‘coming weeks’: Ashwini Vaishnaw

A consultation paper is an official discussion document that presents a possible policy approach and asks affected groups for feedback. It helps the government test ideas before deciding what rules to propose. An AI law, by contrast, would be legislation passed through Parliament and would create binding duties, rights, enforcement powers, or penalties. The paper is therefore a starting point, not the final legal framework.

For example, the paper could ask how humans should control AI agents, models, outcomes, and processes. It might also seek views on a mechanism for stopping systems that go out of control. These topics are mentioned as principles or concerns, but the government has not yet explained how they would become enforceable rules.

The consultation paper is expected in the coming weeks. Vaishnaw said it will seek views from civil society, citizens, industry, and other stakeholders. The article gives no date for a bill reaching Parliament, and no details on its scope, regulator, or penalties.

The proposed law is aimed at harms caused by AI rather than at stopping the technology itself. The article points to risks that have crossed a threshold and says regulation should protect citizens while keeping humans in control. It specifically highlights concerns about autonomous agents, accountability, and AI systems producing harmful outcomes.

A concrete example is an AI agent acting without effective human oversight. The government’s stated principles say control over agents, models, outcomes, and processes must remain in human hands. They also call for an equivalent of a stop mechanism when systems go out of control. The article does not explain who would operate that mechanism or how it would work.

The consultation is expected to explore these unresolved questions. It will also consider harms in sectors where companies are adopting AI rapidly. The government has not yet published the proposed law’s exact scope, enforcement structure, or penalties.

The government previously favoured focused changes to existing laws and voluntary norms instead of a standalone AI statute. India’s AI Governance Guidelines, released on November 5, 2025, followed that approach. In December 2025, ministry secretary S Krishnan said no dedicated AI law was being planned, while a 2023 parliamentary reply said the government was not considering one.

The position changed because officials now say AI risks have crossed a threshold demanding legislation. Critics had argued that existing laws and voluntary measures left gaps, especially around autonomous agents and accountability. A dedicated law could provide a clearer basis for addressing harms, even though its design has not been announced.

Vaishnaw’s planned consultation paper is the first formal step toward a statute. It will gather views from citizens, civil society, industry, and other stakeholders. The government still favours innovation and has not said when a bill might reach Parliament or what regulator and penalties it would establish.

The article presents AI as a technology with broad and fast-growing economic reach. Vaishnaw says banks, manufacturers, insurers, and service firms are adopting it rapidly. This means AI governance cannot focus only on companies that build models. It must also consider organisations that use those systems in everyday operations.

For example, a bank or insurer could use an AI system as part of its services, while a manufacturer or service company could use AI in its processes. The article does not specify the individual tasks involved. Its key point is that users, as well as technology developers, need to help shape the rules because AI errors can affect citizens directly.

Vaishnaw said there is already a good understanding of what AI can deliver and what harm it can do. The consultation will therefore seek input from users and other stakeholders, alongside earlier discussions with technology companies. This wider process reflects AI’s reach across several industries.

When organisations use AI in services that affect people, serious errors can create direct harm for citizens. The article specifically says banking and insurance are areas where AI errors can hit citizens directly. It does not list particular consequences, such as lost money or denied claims, so the safest conclusion is that faulty outputs could damage people’s interests and create accountability problems.

The key mechanism is reliance on AI by a user organisation. A bank, insurer, manufacturer, or service firm may adopt AI rapidly, but responsibility can become unclear if a system produces a harmful result. That is why the article connects user input with questions about accountability and human control.

The proposed consultation could examine how such harms should be regulated and who should remain responsible. Vaishnaw has not explained the regulator, penalties, or detailed scope. The shift toward consulting AI users suggests these real-world effects will be important in the next stage.

Harm-based regulation focuses on the damage or risk an AI system creates. Rules could target unsafe outcomes, weak accountability, or systems operating without sufficient human control. A licensing regime would work differently: every AI system, or a defined class of systems, would need official permission before operating. Licensing therefore controls entry, while harm-based rules focus on conduct and consequences.

The article’s example is an AI agent acting without oversight. A harm-based approach might require human control and a way to stop the system if it goes out of control. A licensing approach might require approval before that agent could be deployed. The article does not say that approval would be required, so this comparison explains the difference rather than describing a confirmed rule.

Vaishnaw said India’s bias has been toward innovation while keeping citizens safe. The article says this line suggests harm-based regulation rather than licensing, but stresses that the government has not spelled out the design. The consultation will help clarify the choice.

AI is difficult for one country to regulate because digital systems do not follow physical borders in the same way traditional activities do. Vaishnaw said that boundaries do not exist in the digital world and that any initiative must be globally aligned. Rules made in one country may therefore interact with technology, companies, users, and effects connected to other countries.

The problem becomes sharper when AI agents act with limited oversight. The article says control over agents, models, outcomes, and processes must remain in human hands, and it raises the need for a way to stop systems that go out of control. If activity crosses borders, deciding who holds that control and which rules apply becomes harder.

Vaishnaw compared the challenge with the nuclear industry, where countries agreed on shared principles. He said India should participate as one of the world’s top three or four AI ecosystems. The consultation is national, but the minister presents international cooperation as essential.