India Spent A Year Saying It Would Not Regulate AI, And Is Now Writing The Paper

The Ministry of Electronics and Information Technology will publish a consultation paper on regulating artificial intelligence within a month, Union Minister Ashwini Vaishnaw said on 8 October, speaking at the India launch of the World Bank's World Development Report 2026.

He said the paper would be regulation-focused and built on four elements: safety, a human-first approach, skilling, and ensuring no person is left behind. He described it as probably the base for the regulation which has to come, a shift from the government's earlier position of relying on existing laws and voluntary compliance.

MeitY secretary S Krishnan said the World Bank report advocates a sequence similar to India's, relying on existing laws, moving to voluntary compliance and considering regulation at a later stage.

The India AI Governance Guidelines, released in November 2025 under the IndiaAI Mission, are a voluntary framework built on seven principles. They create no offences or penalties, establish no AI regulator, and chose not to propose a dedicated AI law. In April 2026 the government set up the AI Governance and Economic Group as a central coordinating mechanism, with membership limited to ministers and senior officials.

MeitY has separately consulted on amendments to the intermediary rules covering synthetically generated information and deepfakes.

Vaishnaw also said the ministry would use an empanelment process rather than tenders for government use of AI, that a roadmap for Indian GPUs for inference is now visible with a solution expected in three to four years, and that the government is procuring 10,000 GPUs for public-interest workloads.

The reversal is the news, and it is worth setting out precisely because the earlier position was held deliberately rather than by default.

In November 2025 MeitY published the India AI Governance Guidelines and explicitly declined to propose a dedicated AI law. The framework was voluntary, principle-based, created no offences and established no regulator. Officials argued consistently that existing statutes could be amended where needed and that AI-specific legislation was unnecessary. The choice was presented as a competitive advantage against the EU's approach, made shortly before India hosts the AI Impact Summit.

Eleven months later the minister responsible describes a forthcoming paper as the base for regulation which has to come. That is not a clarification of the earlier stance; it is a different one.

What changed is not stated, and the candidates are reasonably clear. Deepfakes have moved from a theoretical harm to a live enforcement problem, with the government already consulting on labelling synthetic content. Agentic systems that act rather than answer have arrived in consumer products. And the global position shifted, with the EU's high-risk obligations deferred to December 2027 rather than abandoned.

The sequence India is following, existing law first, voluntary compliance next, regulation when warranted, is defensible and Krishnan notes the World Bank endorses it. The thing to hold the government to is that the sequence was always meant to end here. It is now reaching the final step, and the detail of the paper is what matters rather than the announcement of it.

For companies the useful question is which of them this would actually reach.

India has no frontier lab. The AI being deployed here is overwhelmingly built elsewhere and applied locally, which is the pattern our own reporting keeps finding: $438 million of AI funding last quarter concentrated in four deals, with capital going to the application layer rather than to model training. A safety-focused framework aimed at model developers would land on a handful of foreign companies and very few Indian ones.

A framework aimed at deployers is a different proposition entirely, and far broader. Razorpay is placing brand advertising inside ChatGPT. Insurers are automating underwriting while IRDAI consults separately on distribution and a public registry. Kuku says 60 to 70 per cent of its production uses AI tools. None of those are model developers and all of them make consequential automated decisions.

The deepfake work points towards the second reading, since synthetic media rules bind whoever publishes rather than whoever trained.

Sequencing is the other thing worth watching. Vaishnaw put an Indian GPU roadmap at three to four years away and is procuring 10,000 GPUs in the meantime, which is an admission that India regulates AI while importing nearly all of the compute it runs on. A jurisdiction with limited leverage over supply has to be careful that obligations land on parties it can actually reach, which in practice means Indian deployers rather than foreign developers.

Comments will be the moment to argue about that, and the window opens in weeks.

For daily, sharp analysis of the biggest moves in the Indian business and startup ecosystem, follow StartupFox.