Since 20 February 2026, platforms in India must label AI-generated audio, images and video, and must take down deepfake intimate imagery or impersonation within 2 hours of a complaint. Content flagged by a court or an authorised government officer must go within 3 hours, down from 36. These changes come from amendments to the IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, which MeitY notified on 10 February 2026.

If you are a victim of a deepfake, these deadlines give you real leverage. If you create content with AI, or build AI products, you now have specific duties. This guide explains both.

What counts as “synthetically generated information”

The rules use the term synthetically generated information (SGI). It covers audio, visual or audio-visual content that is created or altered by a computer so that it looks real and shows a person or event in a way that is, or is likely to be, indistinguishable from the real thing. A face-swapped video of a CEO announcing a fake offer is SGI. So is a cloned voice note from a “relative”.

The rules also say what is not SGI:

  • Routine, good-faith editing such as colour correction, compression or formatting that does not change the meaning of the content.
  • Ordinary creation of documents, presentations and training material, as long as they are not false documents or records.
  • Accessibility tools such as translation, captions and image descriptions that do not materially alter the original.

Plain AI-written text is not covered by the labelling rule, which targets audio and visual content. Some early reports said text must also be labelled; that is not what the final rules require.

The key changes at a glance

AreaBeforeAfter 20 February 2026
Removal on court order or authorised government notice (Rule 3(1)(d))36 hours3 hours
Complaints about nudity, sexual content, impersonation or morphed images (Rule 3(2)(b))24 hours2 hours
Other grievances to the platform’s grievance officer15 days7 days
Action on other unlawful or harmful content (Rule 3(1)(b))72 hours36 hours
Labelling of SGINo specific ruleProminent label plus permanent metadata or a provenance marker with a unique identifier
User reminders about the rulesAt least once a yearAt least once every three months

Duties for platforms and AI tool makers

If you offer a tool that creates or edits SGI

  • Label the output prominently, and embed permanent metadata or another provenance mechanism with a unique identifier.
  • Make sure users cannot remove or suppress the label or metadata.
  • Use reasonable technical measures to stop the tool producing child sexual abuse material, non-consensual intimate imagery, false documents or records, content about explosives or weapons, and deceptive impersonation of real people or events.

If you are a large social media platform

Significant social media intermediaries, which are platforms above the government’s user threshold, have extra duties under Rule 4(1A). Before content is published they must ask users to declare whether it is synthetic, use technical tools to check that declaration, and label content that is confirmed as SGI.

What happens if a platform fails

The amendments do not create new fines. The main risk is losing safe harbour under Section 79 of the IT Act, the protection that stops a platform being held liable for what its users post. A platform that knowingly allows unlawful SGI, or fails to act on it, is treated as failing its due diligence. The rules also confirm that removing content in good faith to comply does not cost a platform its safe harbour.

What to do if you are targeted by a deepfake

Speed matters, and the order of steps matters too.

  1. Save evidence first. Take screenshots, copy the URLs and note the account names, dates and times. Do not rely on the content staying online.
  2. Complain to the platform. Use the in-app report option and also write to the grievance officer, whose contact details must be published on the platform. State clearly that the content is a morphed or impersonating image of you. For this category, the 2-hour clock runs from your complaint.
  3. File a cybercrime complaint on cybercrime.gov.in. If money has been lost, call 1930 immediately; reporting within the first hours gives the best chance of freezing the funds.
  4. Escalate if ignored. If the grievance officer does not act, you can appeal to the Grievance Appellate Committee, and you can approach the police or a court.
  5. Tell your employer if the deepfake uses your professional identity, for example to defraud clients. Their security team may need to warn customers.

What it means if you use AI for content

Suppose a marketing team uses an AI video tool to make a product ad with a synthetic presenter. The tool should apply a label and metadata. If the team posts it on a large platform, they will be asked to declare it as synthetic. Stripping the label or ticking “not AI” risks the post being removed and the account being restricted. The safe habit is simple: declare AI use when asked, and never use a real person’s face or voice without written consent.

Light edits such as cropping, colour correction or AI noise removal on a real photo generally fall under the routine-editing exception. Swapping faces, generating a person who appears real, or changing what someone said does not.

If you build AI products

  • Plan for watermarking and provenance metadata from the start. Open standards such as C2PA content credentials are one option.
  • Put abuse filters in front of image, video and voice generation, especially for real names and faces.
  • Make your report-and-takedown process work at weekends. A 2-hour deadline leaves no room for a Monday-morning queue.
  • Document what you do. If a regulator or court asks, you need to show your measures were reasonable.

Spotting deepfake scams

  • Treat video ads showing celebrities or business leaders promising high returns as fake until proven otherwise. Check that any investment platform is registered with SEBI before you put money in.
  • If a “boss” or “relative” calls urgently asking for money, hang up and call back on a number you already have.
  • Agree a family code word for emergencies.
  • Look for mismatched lip movement, odd blinking and flat audio, but do not rely on these alone. Good fakes hide them.

FAQ

Do I have to label a photo I edited with an AI filter?

Usually not, if the edit is routine and does not change the meaning of the image. If the result shows something that did not happen, treat it as synthetic.

Does the 2-hour rule apply to all deepfakes?

No. It applies to complaints about nudity, sexual content, impersonation and morphed images. Other content follows the 36-hour or 3-hour timelines depending on how the platform is notified.

This article is general information, not legal advice. If you are dealing with a specific case, consult a qualified lawyer.

Last reviewed: 8 October 2026.

Sources

Leave a Reply

Trending