India's AI Content Labelling Rules (IT Rules 2026): What Startups That Build or Publish AI Content Must Do

If your startup builds an app that creates images, video or voice with AI, hosts content that users upload, or runs a marketplace, a 2026 change to India's IT Rules affects you. The amendment makes labels and embedded metadata mandatory for AI-generated audio and video, and it sharply shortens the time platforms have to remove unlawful content.

This guide explains what changed, who is covered, and what a startup should do. It is based on published law-firm analyses, listed under Sources. Rule numbers are as reported in those analyses, so please read the Gazette text for your own case.

Quick summary

• The IT (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 were notified on 10 February 2026 (reported as G.S.R. 120(E)) and took effect on 20 February 2026.
• They define "synthetically generated information" (SGI) as audio, visual or audio-visual content created or altered by a computer so that it looks real. Text-only content is not covered.
• Platforms must label such content prominently and, where feasible, embed permanent metadata.
• Large social media platforms must collect user declarations and verify them with technical measures.
• Takedown windows shrink to 3 hours for orders and 2 hours for the highest-risk content.
• Platforms that do not comply can lose safe harbour protection.

What counts as synthetic content

According to published analyses, SGI is audio, visual or audio-visual information created or altered by a computer resource so that it appears real and cannot easily be told apart from a real person or a real event. It covers deepfakes, AI-altered images and videos, and cloned voices. AI-generated text is not SGI, although existing rules on unlawful text still apply.

Three kinds of content are carved out, as long as they do not materially change the substance or create false documents:

• Routine or good-faith editing, such as formatting, colour correction, compression or transcription.
• Routine or good-faith creation of documents, presentations, PDFs, or educational and research material.
• Accessibility improvements, such as clearer audio, translation or searchability.

What platforms must do

Label and tag. Permitted synthetic content must be clearly and prominently labelled. Visual content needs a visual label, and audio content needs an audio disclosure. Where feasible, intermediaries must embed permanent metadata or unique identifiers that let the content be traced to the tool that made it, and they must not allow labels or metadata to be removed or changed. The October 2025 draft's fixed size for the label (10% of the visual area) was dropped from the final rules.

Ask users and verify. Significant social media intermediaries, meaning platforms with more than 5 million registered users in India, must require users to declare whether uploaded content is synthetic, and must verify those declarations using technical measures. They cannot rely on declarations alone. They must display clear labels wherever content is synthetic.

Act quickly. The amendments shorten removal times:

• 3 hours to remove unlawful content after a court order or government notice, down from 36 hours.
• 2 hours for the highest-risk categories, including non-consensual intimate imagery, nudity or sexual acts, impersonation and artificially morphed images, down from 24 hours.
• 36 hours to act on complaints about content concerning an individual, down from 72.
• 7 days for general user grievances, down from 15.

Warn users. Platforms must now remind users every three months, not once a year, of the consequences of breaking the rules, such as account suspension, content removal and penalties. Platforms that let users create synthetic content must warn that violating content can lead to the user's identity being disclosed to victims.

The amendments also replace the earlier "endeavour" wording with a mandatory duty to use "reasonable and appropriate technical measures".

Who is covered

The rules apply to intermediaries, a broad term that includes social media platforms, hosting providers, search engines, marketplaces and others. A platform that lets users create or share synthetic content faces the specific duties for SGI. The largest duties, the user declaration and verification, fall on significant social media intermediaries.

That leaves three kinds of startup:

1. You build or run a platform with user-generated content, a marketplace, a community, or a tool that creates AI images, video or audio. You are likely an intermediary, and you should assume the labelling, metadata and takedown duties apply.
2. You only use AI tools to make marketing content. You are probably not an intermediary, so the platform duties do not apply to you directly. But the platforms you post on will label or remove synthetic content, and misleading synthetic content can create legal and reputational risk.
3. You provide AI tools to businesses. Your customers who run platforms will ask you for labelling and metadata features in the tool.

Published analyses do not separately address ordinary businesses that are not intermediaries, so take advice if you are unsure which group you fall into.

A further advisory

A MediaNama report describes an advisory from MeitY dated 8 October 2026 that asks platforms to label synthetic content presented as current events and to prioritise official complaints. As reported, it applies to platforms, not directly to businesses that publish content on them, and its labelling clause does not limit itself to audio-visual content. Check its status in the primary source.

Open questions

Law firms point out that the rules leave open what counts as "reasonable and appropriate technical measures", how much effort is expected to detect AI content, and how a two-hour takedown works in practice for a small team. Keep records of the steps you take, because a platform that can show a careful process is in a better position if its decisions are questioned.

A checklist for startups

1. Decide whether you are an intermediary. Look at what users can upload, generate or share on your product.
2. Add labels and metadata to anything your tool generates as audio or video, and make sure users cannot remove them.
3. Build a declaration step where users upload content, and think about how you would verify it.
4. Write a takedown runbook. Name who receives notices, how to reach them around the clock, and how to act within 2 to 3 hours.
5. Check your grievance process against the 36-hour and 7-day timelines.
6. Update your terms and user notices, and send the reminder every three months.
7. Review vendor contracts with AI model providers, so labelling and metadata features, and responsibility for them, are clear.
8. Label your own AI content voluntarily, especially anything that could be mistaken for a real person or event.

Frequently asked questions

Do these rules apply to AI-written text?

No. SGI covers audio, visual and audio-visual content. Existing rules on unlawful content still cover text.

Is a small startup treated like a large platform?

The user declaration and verification duties fall on significant social media intermediaries (over 5 million registered users). Other intermediaries face the labelling, metadata and takedown duties.

What happens if we miss a takedown deadline?

A platform that fails its due diligence duties can lose safe harbour, which protects it from liability for what users post.

Is editing a photo in an app covered?

Routine or good-faith editing, such as colour correction, is carved out, as long as it does not materially change the content.

This article is general information and not legal advice. The rules and advisories are changing, so confirm the position for your product with a lawyer.

Building a product with user content or generative AI? Talk to the StartupKaro team about compliance for your launch.

Sources

• Freshfields: India targets deepfakes and AI-generated content, key changes under MeitY's 2026 amendments to the IT Rules
• Lexology: India's 2026 IT Rules amendment, regulating AI-generated content and accelerating compliance
• iPleaders: IT Rules 2026
• Mondaq: India tightens oversight on AI-generated content under IT Rules
• HLC: India introduces mandatory labelling for AI and 3-hour takedown for illegal content
• Veritect: IT Rules 2026 synthetic media amendment, labelling, metadata, 3-hour takedown
• MediaNama: MeitY widens labelling for social media beyond synthetic content