On 10 February 2026, MeitY notified the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, and they took effect on 20 February 2026. They are India's first binding rules aimed specifically at AI-generated media, and they change obligations for three groups at once: the AI tools that generate content, the social platforms that host it, and the people and businesses who post it.
This post explains what the notified rules require, how the final version differs from the October 2025 draft, and what a business or creator using AI video, image or voice tools should do differently.
What counts as "synthetically generated information"
The rules introduce a defined term, synthetically generated information (SGI): audio, visual or audio-visual content created or altered artificially or algorithmically in a way that makes it appear real — a person, voice or event that looks or sounds authentic.
Three exclusions matter in practice. AI-generated text on its own is not SGI, although existing rules on unlawful content still apply to it. Routine or good-faith editing — colour correction, noise reduction, cropping and similar — is outside the definition. So are accessibility and quality tools such as translation, captioning and descriptions that don't manipulate the underlying content. A brightened product photo isn't SGI; an AI-generated spokesperson who doesn't exist is.
If you run an AI generation tool
Any intermediary that lets users create or alter content with AI now has to label SGI clearly and prominently — a visible label on images and video, and a spoken disclosure at the start of audio. Where technically feasible, it must also embed permanent metadata or another provenance mechanism, including a unique identifier that traces the content to the tool used to make it, and must not let users remove or alter the label or metadata.
Such tools also have to use automated safeguards to stop users generating prohibited categories — child sexual abuse material, non-consensual intimate imagery, obscene content, forged documents, material that helps procure weapons, and deceptive depictions of real people or events — and must remind users of the consequences every three months rather than once a year.
The October 2025 draft would have required visible labels covering at least 10% of an image's surface, or the first 10% of an audio clip's duration. That was dropped after industry pushback. The final rules only say labels must be clear and prominent, leaving the method to each platform.
If you run a large social platform
Significant social media intermediaries — the large platforms — must ask users to declare whether an upload is AI-generated, use technical measures to check that declaration rather than just trusting it, and label content confirmed as SGI. A platform that knowingly allows unlawful SGI is treated as failing its due-diligence duties, which puts its safe-harbour protection from liability for user content at risk.
The amendment also sharply shortened response times. Takedowns ordered by a court or the government now have to happen within 3 hours instead of 36. Non-consensual intimate imagery, impersonation and morphed images have to come down within 2 hours instead of 24. General user grievances must be resolved in 7 days instead of 15.
What it means if you make content with AI
Most of the legal duties fall on platforms and tool makers, not on the individual creator. But they reach you in practice, in four ways.
Expect to declare. When you upload an AI-generated video, ad or voiceover to a large platform, you'll increasingly be asked whether it's AI-generated. Answer truthfully: a false declaration breaks platform terms and risks the content being removed or your account being restricted, and the platform is required to check.
Expect labels you can't remove. Output from compliant AI tools will carry visible labels and embedded metadata. Stripping them to make content look real is exactly what the rules target.
Be careful with real people. AI avatars of real people, cloned voices and synthetic endorsements are the riskiest category. Get written consent from anyone whose face or voice you recreate, and don't depict real people saying or doing things they didn't. With 2-hour takedown windows, a complaint can remove a campaign almost immediately.
Brief your agencies. If an agency or freelancer produces AI content for your brand, the rules apply to that content wherever it's published. Put labelling and consent requirements in the brief.
How the tools in our catalogue fit
The AI video, image and voice tools we list — HeyGen, Synthesia, Runway, Adobe Firefly, Midjourney, ElevenLabs, Murf and others — are the kind of tools the SGI provisions are aimed at. How each one labels output, embeds provenance metadata (some already use the C2PA content-credentials standard) and restricts real-person likenesses varies by product and plan, and is changing as they adapt to India's rules. Check the current export settings in whichever tool you use rather than assuming.
What's still unclear
Three questions are still open: how strictly "technically feasible" will be read for embedding metadata, particularly for smaller or open-source tools; how platforms will verify user declarations at scale without false positives on edited but real footage; and how the rules will be enforced against overseas tools without an Indian presence. This post is a plain-language summary based on the notified rules and law-firm analyses of them, not legal advice.