India now requires AI media to label itself.

While much of the world argues about AI transparency, India quietly made it binding. Its amended Information Technology Rules have been in force since February 20, 2026, and they place real duties on anyone whose service can create or spread realistic synthetic audio or video. The instrument carries no “AI” in its title, which is why it went largely unnoticed abroad, but its reach is broad and its enforcement mechanism has teeth.

What “synthetically generated information” means

SGI is audio, visual, or audio-visual content that is artificially or algorithmically created or altered so that it appears real and depicts a person or event in a way likely to be perceived as indistinguishable from reality. One boundary matters enormously for global products: text is not SGI here. A chatbot’s false article is outside this definition, though it may be unlawful on other grounds. That is the opposite of the EU’s approach.

The three duties

• Prevent unlawful SGI. Deploy reasonable technical measures, including automated tools, against categories such as non-consensual intimate imagery, child sexual-abuse material, false documents, and deceptive impersonation of a real person or event.
• Label permitted SGI, clearly and prominently. A visible label on visual content and a spoken disclosure prefixed to audio, plus embedded permanent provenance metadata with a unique identifier that marks the content as SGI and identifies the tool that made it.
• Block tampering. The service must not offer a way to modify, suppress, or remove the label, metadata, or identifier. In the regulator’s own words, no “remove watermark” or “export without metadata” feature.

Large platforms carry more

Significant social-media intermediaries, defined as those with more than five million registered users in India, must go further. Before content is displayed, they must obtain a user declaration on whether an upload is SGI, deploy measures to verify that declaration, and label confirmed SGI. A platform that knowingly permits or fails to act on non-compliant SGI is deemed to have failed its due diligence. The standard also tightened across the board: “endeavour to deploy” became simply “deploy.” Best-efforts is gone.

The sanction, and two myths to drop

This is not a fine schedule. Non-compliance is a failure of due diligence, which forfeits safe harbour under Section 79 of the IT Act — exposing the intermediary to liability for third-party content, plus penalties under the Act and general law. Two myths are worth killing before they cost you engineering time:

• There is no “10% of the screen” watermark rule. That quantitative mandate appeared in an October 2025 draft and was removed from the notified Rules, replaced by a qualitative “clear and prominent” standard. Anyone told to build a 10% watermark is complying with a rule that never took effect.
• Text-only AI output is not covered. Only audio and visual SGI are in scope. This is the single most consequential difference from the EU for anyone shipping into both markets.

An honest limitation

This summary is drawn from the Ministry of Electronics and Information Technology’s official FAQ and the gazette citation for the amended Rules. The FAQ is not itself the law, so verify the enacted gazette text before acting on any specific clause, and expect enforcement practice to develop. The durable point holds regardless: labeling, provenance, and anti-tampering for synthetic media are moving from aspiration to obligation, and India is now one of the largest markets to make them binding.

This briefing is general information from Sentinel Assurance Group, not legal advice. Regulatory dates and requirements change — we maintain these briefings, but verify against primary sources and counsel before acting. Last reviewed August 10, 2026.