AI-generated content labelling rules in India for app builders

If your product generates images, video or voice, India now has a rule about how that output must look and sound to the person receiving it. MeitY amended the IT Rules, 2021 on 10 February 2026, and its FAQ on the amendment is detailed enough to build from. This page summarises what the rules call "synthetically generated information", who has to label it, what a label must contain, and which clocks changed. It is written for product managers and founders, not as legal advice.

1. What changed and when

MeitY notified the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 as G.S.R. 120(E), dated 10 February 2026. They came into force on 20 February 2026.

A further draft is out. On 30 March 2026 MeitY published draft amendments, and on 21 April 2026 it said it was adding a change to the label rule that requires "continuous and clearly visible display of label for synthetically generated information throughout the duration of the content in visual display". Comments were due by 7 May 2026. I did not find a final notification, so check MeitY's site for its status before you design around it.

2. What counts as synthetically generated information

The rules define synthetically generated information (SGI) as audio, visual or audio-visual information that is artificially or algorithmically created or altered so that it appears real, and that depicts a person or an event in a way that is, or is likely to be perceived as, indistinguishable from a natural person or a real-world event.

MeitY's examples of SGI include:

• a realistic AI video of a person speaking as if on camera;
• synthetic voice, whether text-to-speech or voice cloning, that reproduces a natural person's voice with realistic intonation and tone;
• an AI reconstruction of an accident or crowd scene made to look like real footage;
• a realistic AI image of a person at an event.

Three things are outside the definition:

• Routine, good-faith editing: brightness, compression, noise reduction, transcription, stabilising.
• Routine creation of documents, slides, training material and research outputs, as long as it does not produce false documents or records.
• Tools used only for accessibility, translation or searchability, such as captions, audio description or auto-generated summaries and tags.

Plain text is not SGI. MeitY's FAQ says pure text outputs by themselves are not SGI because the definition is limited to audio, visual and audio-visual information. A chatbot article that spreads a false rumour, with no synthetic media, is not SGI, though it can still be unlawful information under other rules.

3. Who must comply

The new due-diligence rule applies to an intermediary that offers a computer resource which enables or facilitates creating, generating, modifying, publishing, sharing or spreading SGI. MeitY lists AI image and video generation or editing tools, voice synthesis and cloning tools, and platforms that carry SGI. Its example: a tool that generates realistic AI videos or cloned voice notes is covered.

The FAQ does not say how a thin app built on someone else's model API is treated. If your product lets a user generate realistic people, voices or events, plan as if you are covered, and ask a lawyer.

4. What you have to do

Block the prohibited kinds. Covered intermediaries must deploy reasonable and appropriate technical measures, including automated tools, so users cannot create or share SGI that breaks the law. The rules name: child sexual abuse material and non-consensual intimate imagery; false documents or electronic records; content on making explosives, arms or ammunition; and content that falsely depicts a real person or event in a way likely to deceive, including a false identity, voice or statement.

Label what is allowed. For other SGI the rules require two things:

• A clear and prominent label or notice that the content is synthetic. For visual and audio-visual content the label must be prominently visible. For audio there must be a prominently prefixed audio disclosure. MeitY's example is a spoken line at the start such as "This audio is synthetically generated".
• Permanent metadata or other technical provenance, to the extent technically feasible, including a unique identifier, so the content can be identified as SGI and the computer resource that made it can be identified.

The FAQ includes a lawful audio message made with text-to-speech as an example that needs the prefixed disclosure. If you build voice products, read that example closely.

Do not let users strip the label. An intermediary must not enable modification, suppression or removal of the label or the embedded metadata. MeitY's example of what to avoid is a "remove watermark" or "export without metadata" feature.

Warn users. An intermediary whose tool enables SGI must warn users that unlawful SGI can attract penalties under the IT Act and other laws. The warning should be prominent, including at onboarding and periodically. The general reminder to users now goes out at least once every three months; before it was once a year.

Large platforms do more. Significant social media intermediaries, which PIB describes as those with 50 lakh or more registered users in India, must get a user declaration on whether content is SGI before publication and use technical measures to check it. Confirmed SGI must carry a clear label.

5. The clocks that moved

The old values come from PIB's August 2026 summary of the amendment and MeitY's FAQ. The FAQ also says that removing content in line with the rules, including through automated tools, does not by itself cost the intermediary its exemption under section 79 of the IT Act.

6. A build checklist for a generation feature

• [ ] Decide for each output type whether it can look like a real person or event. If yes, treat it as SGI.
• [ ] Add classifiers or rules that refuse the prohibited kinds, and log what they block.
• [ ] Add a visible label for images and video and a spoken prefix for audio.
• [ ] Write provenance metadata with a unique identifier into every file, and keep it in exports and downloads.
• [ ] Remove any "no watermark" or "strip metadata" option.
• [ ] Show the user warning at sign-up and re-show it every three months.
• [ ] Set up a round-the-clock contact route that can act inside the new takedown and grievance periods.
• [ ] Re-read MeitY's site before launch for the status of the draft continuous-label change.

Check with a lawyer how these rules apply to your product before you rely on this page.

Questions people ask

Do AI-written articles or chatbot replies need an AI label?

Not under the SGI rules. MeitY's FAQ says text-only outputs are not SGI by themselves. They remain subject to the general rules on unlawful information.

Does an AI voice-over need a disclosure?

MeitY's FAQ says a lawful synthetic audio message, for example a text-to-speech announcement, should carry a prominently prefixed audio disclosure and embedded metadata. Whether a particular voice falls inside the definition turns on whether it appears real and depicts a person in a way that is indistinguishable from a natural person.

Is the label size fixed?

The rules in force require prominent visibility, and MeitY's FAQ gives no size. A March 2026 draft would add that the label stay visible for the full duration of visual content; I found no final version.

What happens to a platform that ignores the rules?

An SSMI that knowingly permits, promotes or fails to act on SGI in breach of the rules is deemed to have failed to exercise due diligence under the IT Rules, per MeitY's FAQ.

Sources

1. Frequently asked questions on the IT (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, MeitY, 10 February 2026
2. Government Strengthens Regulatory Framework to Address AI-Generated Deepfakes, PIB, 6 August 2026
3. Notice on extension of time for feedback on draft IT Rules amendments, MeitY, 21 April 2026

Related on this site: AI safety for product managers and the ElevenLabs voice AI guide.