In short

  • No AI statute exists, and MeitY's position is that none is needed yet.
  • Binding since 20 February 2026: labelling and metadata for AI-generated audio and video under the IT Rules.
  • The DPDP Act applies to personal data in training sets, with a narrow exclusion for data the person published themselves.
  • Copyright in training data is unsettled. ANI v OpenAI has produced only an interim order.

A founder building on a large language model in India usually asks one question: which AI law do we have to comply with? The honest answer is that there is no single one. The rules that bind come from laws written before generative AI, plus one set of amendments written specifically for it.

This guide separates what is binding from what is guidance, and both from what is still being argued in court. Where a rule has to be built into a product, it says what the engineering team will need to do.

No AI Act, by design

India has not passed an AI statute, and the Government's published position is that it does not need one yet. The India AI Governance Guidelines released by MeitY on 5 November 2025 say that "at this stage, a separate law to regulate AI is not needed given the current assessment of risks."

That makes India different from the European Union, which regulates AI systems by risk category under its own AI Act. In India the question is always which existing law the AI product touches: the Information Technology Act, 2000 and its rules for platforms, the DPDP Act, 2023 for personal data, the Copyright Act, 1957 for training material and outputs, and sector rules where a regulator has spoken.

The deepfake rules: binding since February 2026

The one set of rules written for generative AI is the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, notified as G.S.R. 120(E) on 10 February 2026 and in force from 20 February 2026. They amend the IT Rules, 2021 and apply to intermediaries, which includes platforms that let users create or share content.

What counts as synthetic

Rule 2(1)(wa) defines "synthetically generated information" as audio, visual or audio-visual information "artificially or algorithmically created, generated, modified or altered using a computer resource" so that it appears real and depicts a person or event in a way likely to be perceived as indistinguishable from a natural person or real-world event.

Three things fall outside it: routine or good-faith editing such as colour correction, noise reduction, transcription or compression; good-faith documents, presentations, educational and research material; and uses for accessibility, translation or searchability. Text is not covered, because the definition is limited to audio and visual information.

Labels and metadata

Under Rule 3(3)(a)(ii), an intermediary whose tools create such content must make sure it is "prominently labelled", visibly for images and video and by a prefixed audio disclosure for sound. It must also be "embedded with a permanent metadata or other appropriate technical provenance mechanisms, to the extent technically feasible, including a unique identifier" identifying the intermediary's computer resource. Rule 3(3)(b) bars the intermediary from enabling anyone to modify, suppress or remove the label or metadata.

The October 2025 draft required a label covering at least ten per cent of the screen. The final rules dropped that figure and require prominence instead.

What must be blocked

Rule 3(3)(a)(i) requires technical measures to prevent synthetic content that is child sexual abuse material, non-consensual intimate imagery or obscene; that creates false documents or electronic records; that relates to explosives or arms; or that "falsely depicts or portrays a natural person or real-world event" in a way likely to deceive.

Extra duties for large platforms

Under Rule 4(1A), a significant social media intermediary must, before content goes live, ask users to declare whether it is synthetic, use technical measures to verify that declaration, and label content confirmed as synthetic. A platform that knowingly permits or promotes violating content is deemed to have failed its due diligence, which puts its safe harbour under Section 79 of the IT Act at risk.

Faster takedowns

The same amendment shortened response times across the IT Rules. Content must now be taken down within 3 hours of a court order or government intimation, down from 36. Complaints about intimate imagery must be acted on within 2 hours, down from 24. These apply to all content, AI-generated or not.

A further draft amendment, published for comment in March and April 2026, proposes that labels stay visible for the full duration of synthetic content. As far as I can find, it has not been notified.

Personal data and AI

The DPDP Act, 2023 does not mention AI, but it applies to any processing of digital personal data, which includes collecting it for a training set, fine-tuning on customer records, and running inference on user inputs. Its main duties take effect on 13 May 2027.

The exclusion AI companies most often rely on is Section 3(c)(ii). The Act does not apply to personal data "made or caused to be made publicly available" by the Data Principal it relates to, or by a person under a legal obligation to publish it. The Act's own illustration is a person who shares her views and personal data publicly on social media.

The exclusion is narrower than "anything on the internet". Data posted about a person by someone else, leaked data, and data the person shared with a limited audience are not covered by it. A scraped dataset will usually contain all three.

For a business that the Government notifies as a Significant Data Fiduciary, Rule 13(3) of the DPDP Rules adds a specific duty: due diligence to verify that "algorithmic software" handling personal data is not likely to pose a risk to Data Principals' rights. See the Significant Data Fiduciary guide.

The leading Indian case is ANI Media Pvt Ltd v. OpenAI OpCo LLC in the Delhi High Court. On 24 July 2026 the court dismissed ANI's application for an interim injunction. It took the "prima facie view that Open AI's act of storing ANI's original literary works for training LLMs underlying ChatGPT falls under Section 52(1)(a) of the Copyright Act" (the fair dealing exception) and so was not infringement.

Two cautions. First, this is an interim order. The court said its observations "would have no bearing on the final outcome of the suit", which is still pending. Second, a fair dealing finding depends on the facts, so the order does not settle the position for every model or every kind of content.

Separately, DPIIT published a "Working Paper on Generative AI and Copyright, Part 1" in December 2025. It proposes a hybrid model in which AI developers get a blanket licence to train on lawfully accessed content, with royalties payable only once the AI tool is commercialised, at rates set by a government-appointed committee. It is a consultation paper. The Copyright Act has not been amended.

Guidance that is not law

These documents shape how regulators think, and are worth following, but none of them creates legal obligations by itself:

  • India AI Governance Guidelines (MeitY, 5 November 2025). Seven principles, called sutras: trust, people first, innovation over restraint, fairness and equity, accountability, understandable by design, and safety, resilience and sustainability. The document describes its approach as supported by voluntary measures that are "not legally binding".
  • RBI FREE-AI report (13 August 2025). A committee report setting out a framework for responsible AI in the financial sector, with 26 recommendations. It is not a direction to regulated entities.
  • MeitY advisories to intermediaries, including those of 15 March 2024 and 16 March 2026 on synthetic content. Advisories have no statutory force of their own. On labelling, the binding Rule 3(3) now does the work.

A working checklist

  • If your product lets users generate images, video or audio, build labelling and metadata into the output pipeline, and make sure users cannot strip them.
  • Keep a record of where each training dataset came from, and whether any personal data in it falls within Section 3(c)(ii) or needs a lawful basis under the DPDP Act.
  • If you fine-tune on customer data, check that your privacy notice and contracts cover it.
  • Track ANI v OpenAI and the DPIIT working paper before relying on fair dealing for commercial training.
  • If you are a financial or securities business, read your regulator's AI documents alongside this list.

Frequently asked questions

No. As of November 2026 there is no AI-specific statute. MeitY's India AI Governance Guidelines of 5 November 2025 state that a separate law to regulate AI is not needed at this stage. AI products are governed by existing law, chiefly the Information Technology Act, 2000 and its rules, the DPDP Act, 2023, and the Copyright Act, 1957.

Yes, for intermediaries whose tools create synthetically generated audio, visual or audio-visual content. Since 20 February 2026, Rule 3(3) of the IT Rules, 2021, as amended by G.S.R. 120(E), requires such content to be prominently labelled and embedded with permanent metadata or another provenance mechanism, to the extent technically feasible.

No. The definition of synthetically generated information in Rule 2(1)(wa) covers audio, visual and audio-visual information only, and excludes routine editing, good-faith documents and accessibility uses.

Not settled. In ANI v. OpenAI (24 July 2026) the Delhi High Court took the prima facie view, at the interim stage, that storing ANI's works to train the models behind ChatGPT fell within Section 52(1)(a) of the Copyright Act and was not infringement. The court said its observations have no bearing on the final outcome, and the suit continues.

Section 3(c)(ii) says the DPDP Act does not apply to personal data made publicly available by the Data Principal, or by a person legally obliged to publish it. Data that was published by someone else, leaked, or made public without such an obligation is not covered by this exclusion.


Related reading

Enrolment: Bar Council of Punjab & Haryana  |  No.: PH/9616/2023
This page is informational and does not constitute legal advice. AI regulation in India is changing quickly; the position stated is as of the date shown above.