India eases AI model launch plan after industry backlash

India has revised a recent AI advisory and removed language asking firms to get government approval before launching or deploying AI models to users in the South Asian market. The updated guidance instead focuses on labels, user notices, and clearer identification of unreliable AI output, deepfakes, and misinformation.

WTF Index NEUTRAL
◄ Terminator 1 Idiocracy 1 ►

The story is mainly a regulatory update about disclosure for unreliable AI output and deepfakes, with no strong drift toward either dystopian pole.

India eases AI model launch plan after industry backlash

India has softened a recent AI advisory after criticism from local and global entrepreneurs and investors. The updated version no longer asks companies to obtain government approval before launching or deploying an AI model to users in the South Asian market.

The change shifts the focus from pre-launch permission to user disclosure. Under the revised guidelines, firms are advised to label under-tested and unreliable AI models so users understand that the systems may be fallible or unreliable.

What changed in the advisory

The earlier advisory, dated March 1, asked firms to take government approval before releasing or deploying AI models to users. That requirement is now gone from the version shared with industry stakeholders on Friday.

Instead, the revised advisory asks companies to be clearer with users when an AI model has not been fully tested or may produce unreliable output. The practical emphasis is on disclosure at the point of use, not permission before launch.

Intermediaries are also advised to use "consent popups" or similar mechanisms. The purpose is to explicitly inform users about the unreliability of AI-generated output before they rely on it.

Why the revision matters

The rollback follows severe criticism received by India’s IT ministry earlier this month. The response came from many high-profile individuals, including Martin Casado, a partner at venture firm Andreessen Horowitz, who called India’s move "a travesty."

The March 1 advisory was notable because it marked a reversal from India’s previous hands-off approach to AI regulation. Less than a year ago, the ministry had declined to regulate AI growth and identified the sector as vital to India’s strategic interests.

The latest version does not return the debate to a fully hands-off position. It keeps several expectations for AI companies and intermediaries, especially around unlawful content, user warnings, and the risks of misleading synthetic media.

Rules still emphasized

The advisory continues to stress that AI models should not be used to share unlawful content under Indian law. It also says AI systems should not permit bias, discrimination, or threats to the integrity of the electoral process.

Those points keep the government’s attention on the effects of AI output, even as it drops the approval requirement for launching models. The result is a softer process for deployment, but not an absence of expectations for companies operating in the market.

The advisory also keeps its focus on deepfakes and misinformation. Intermediaries are advised to make such content easily identifiable by labeling it or embedding it with unique metadata or identifiers.

  • No approval requirement: the revised version no longer asks firms to get government approval before launch or deployment.
  • Reliability disclosures: under-tested and unreliable AI models should be labeled for users.
  • User notices: intermediaries are advised to use "consent popups" or similar mechanisms for AI-generated output.
  • Deepfake identification: labels, unique metadata, or identifiers remain part of the guidance.

What was removed

One notable deletion concerns message traceability. The ministry no longer requires firms to devise a technique to identify the "originator" of any particular message.

That change narrows the advisory’s demands while preserving its emphasis on labeling and identification. Rather than requiring a method to trace a message back to its originator, the revised guidance points companies toward making AI-generated or manipulated content easier to recognize.

The new advisory, like the original version from earlier this month, has not been published online. TechCrunch reviewed a copy of it.

A signal about future AI regulation

Earlier this month, the ministry said the advisory was not legally binding. At the same time, it said the advisory signals the "future of regulation" and that the government required compliance.

That framing makes the revision important beyond the immediate wording. The government has stepped back from a pre-approval approach, but the advisory still shows the areas where officials are concentrating attention: unlawful content, unreliable AI output, bias, discrimination, electoral integrity, deepfakes, and misinformation.

For AI firms, the clearest takeaway from the revised guidance is that launching a model is no longer framed around prior government approval. The burden shifts toward informing users when systems are unreliable and making synthetic or misleading content easier to identify.