India’s current deepfake framework is built around an amendment to the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021. Notified on 10 February 2026 and brought into force on 20 February, the amendment creates a framework for “synthetically generated information” (SGI), covering realistic computer-created or altered audio, visual and audio-visual material. It adds prominent labelling and provenance duties, tighter platform safeguards and faster takedown timelines for specified legal notices. [2] [3]
Key points
- India’s 2026 measure is an amendment to the IT Rules, not a standalone AI Act. [2] [3]
- SGI covers realistic synthetic audio, visual and audio-visual material that can appear indistinguishable from a person or real-world event. [2]
- Permissible SGI generally needs a prominent label or audio disclosure plus permanent metadata or another feasible provenance mechanism. [3]
- The three-hour removal window applies to unlawful information after actual knowledge through a valid government intimation or court order; it is not an automatic deadline for every complaint. [1] [4]
- Significant Social Media Intermediaries must require user declarations, verify them with technical measures and label confirmed SGI. [3]
What happened
The legal change came through the Gazette notification for the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026. The notification is dated 10 February and states that the rules would commence on 20 February. It adds a definition of SGI to the existing intermediary framework rather than creating a separate statute devoted only to artificial intelligence. [2]
Under the definition, SGI is audio, visual or audio-visual information created, generated, modified or altered with a computer resource in a way that makes it appear real, authentic or true. The material must depict a person or event in a manner that is, or is likely to be perceived as, indistinguishable from a natural person or real-world event. [2]
The amendment also says what does not count. Routine or good-faith editing, formatting, enhancement, technical correction, colour adjustment, noise reduction, transcription and compression are excluded when they do not materially alter or misrepresent the underlying meaning. Accessibility, clarity, translation and discoverability improvements are also excluded when they do not manipulate a material part of the source. [2] This distinction matters because a rule aimed at deceptive synthetic media is not written as a ban on ordinary post-production or accessibility tools.
What platforms must do
The amended Rules place a direct due-diligence obligation on an intermediary that enables SGI to be created, generated, modified, published, shared or disseminated. It must deploy reasonable and appropriate technical measures, including automated tools or other suitable mechanisms, to prevent SGI that violates applicable law. The listed categories include child sexual exploitation and abuse material, non-consensual intimate imagery, obscene or sexually explicit material, false documents or electronic records, material linked to explosives or weapons, and deceptive portrayals of a person or real-world event. [3]
The Rules distinguish prohibited or unlawful SGI from permissible synthetic material. For SGI that is not covered by the prohibited categories, the intermediary must provide a prominent visual label or a prefixed audio disclosure that lets users identify it as synthetically generated. The content must also carry permanent metadata or another appropriate technical provenance mechanism, to the extent technically feasible, including a unique identifier for the intermediary’s computer resource used to create or alter it. Platforms must not enable those labels or provenance markers to be removed or suppressed. [3]
The government’s July 2026 account describes the same approach as a combination of clear labelling, traceable metadata, technical measures and user accountability. It says the wider framework also relies on existing provisions of the IT Act and the Bharatiya Nyaya Sanhita for offences such as impersonation, privacy violations, unlawful publication and certain forms of public mischief. Those criminal-law routes are separate from the platform due-diligence duties and should not be presented as a single new “deepfake offence.” [1]
Why it matters
The amendment shifts part of the compliance burden from reacting to complaints toward preparing systems in advance. A platform that offers tools for generating or sharing synthetic media must build processes to screen unlawful categories and preserve a way for users to recognise permissible synthetic content. For large social platforms, the framework adds a further check before publication.
A Significant Social Media Intermediary must require a user to declare whether uploaded information is SGI. It must use appropriate technical measures to verify the declaration, taking account of the content’s nature, format and source. If the declaration or verification confirms that the material is synthetic, the SSMI must prominently display a label or notice and cannot publish it without the declaration or label. [3]
The timing rule is equally important but easy to overstate. The government says the removal period for unlawful information was reduced from 36 hours to three hours when an intermediary gains actual knowledge through a valid, reasoned intimation from the appropriate government or a court order. The Indian Express reports the same court- or law-enforcement-directed route and notes that the shorter window requires rapid response and moderation capacity. [1] [4] That does not mean every report by a user, every suspected manipulation or every disputed post is automatically subject to a three-hour clock.
What remains uncertain
The text establishes duties, but it does not by itself show how consistently platforms are applying them or how often authorities are using the faster route. No current implementation audit or enforcement dataset was located for this draft. The official MeitY FAQ explains the amendment but expressly says it is not a legal document; the Gazette and consolidated Rules remain the controlling texts. [3]
Questions also remain about false positives, satire, parody, edited journalism and the reliability of automated detection. Legal analysis has raised concerns about operational ambiguity, enforcement errors, chilling effects and the need to reconcile rapid removal with constitutional speech safeguards. [5] Those are attributed policy and legal concerns, not findings that the Rules have already been struck down or suspended.
It is also important not to recycle the numbers from an earlier proposal as though they were current law. Reuters reported in October 2025 that a draft proposal contemplated labels covering at least 10% of a visual display or the opening 10% of an audio clip and invited comments. That report describes a proposal; the notified February 2026 Rules are the current legal reference and instead require prominent labelling and provenance to the extent technically feasible. [6] [2]
What happens next
The practical test will be implementation. Platforms will need to document how they classify SGI, verify declarations, attach or preserve provenance markers and respond to valid notices within the applicable time. Users and creators will need clearer information about when synthetic content is permitted, when it must be labelled and when conduct may trigger separate legal consequences.
For readers, the safest interpretation is precise rather than absolute: India now has a notified SGI framework inside its intermediary rules, with stronger technical and disclosure duties. It is not accurate to describe the measure as a blanket ban on all AI-generated content, nor to treat the three-hour route as a universal takedown promise. Any later court ruling, amendment, official implementation guidance or major platform compliance change would warrant a fresh update.
