What Indian law asks of a team that collects web data, and how to show that you answered it.

Indian law gives no single rule for AI and web data.

It has no text and data mining exception and no opt-out statute. It has a law on access to computer systems that turns on permission, one interim court ruling on AI training, a government proposal on licensing and a data protection law that is coming into force in stages. This page explains each in plain language, says what is settled and what is open, and shows how RightSignal helps you collect in good faith and keep the evidence. It describes RightSignal's understanding on 10 November 2026 and is not legal advice.

Request accessTalk to us

Section 43 asks one question, and it is about permission

Section 43 of the Information Technology Act, 2000 applies where a person acts "without permission of the owner or any other person who is incharge of a computer, computer system or computer network". Two of the acts it lists matter to anyone who runs a crawler. The first is accessing the system. The second is downloading, copying or extracting data from it. A person who does either without permission is liable to pay damages by way of compensation to the person affected. Section 66 makes the same acts a criminal offence where they are done dishonestly or fraudulently, with a penalty of up to three years in prison, a fine of up to five lakh rupees, or both.

Section 43, Information Technology Act, 2000

without permission of the owner or any other person who is incharge of a computer, computer system or computer network

Three features of the section deserve a business reader's attention.

  • The test is permission. The section asks whether the owner allowed what you did.
  • The words of the section do not ask whether a password was broken or a paywall was passed. On the text, the absence of a technical barrier is not an answer.
  • The section does not say how an owner must express permission or refusal. A site can speak through robots.txt, through its terms or through a barrier.

No Indian court has yet ruled on whether crawling a public web page against a site's stated wishes is access without permission. The government has said that the Act reaches scraping. In February 2025 the Minister of State for Electronics and Information Technology answered a question on the subject in the Rajya Sabha. The answer said that scraping of publicly available user data by intermediaries, including to train AI models, is regulated under the Information Technology Act and its rules, and it referred to Section 43.

When the question is open and the test is permission, the cautious course is to look in the places where a site states its permission and to respect what you find there. Reading a robots.txt rule, a written refusal or a barrier as a withdrawal of permission is RightSignal's policy. It is not a statement of decided law. It is the reading that leaves you with the least to explain later.

Indian websites already invoke Section 43

Indian websites are not waiting for the courts. News publishers, commercial platforms and government portals cite Section 43 in their terms to reinforce limits on access to their data and on extraction from it. RightSignal reviewed a range of these policies in October 2026 and found a consistent pattern.

Most operators cite Section 43 for the way data is obtained. Some go further and name it as a possible basis for action against unlicensed AI training and other later uses.

example.in/terms-of-use

Automated access to this website, including scraping and the use of its content to train artificial intelligence models, is not permitted without a licence. Unauthorised access or extraction may be pursued under Section 43 of the Information Technology Act, 2000.
Invented example

Across the policies reviewed, operators use their terms to spell out which uses they permit, which need a licence and which technical controls a visitor must respect. They cite Section 43 as a basis for seeking compensation for unauthorised access or extraction. Where an operator also cites Section 66, criminal liability under that section requires dishonest or fraudulent conduct.

A site cannot make conduct an offence by citing a statute, and whether Section 43 applies in a given case is a question for a court. A clause of this kind still tells you in writing that the owner does not permit what you plan to do, and permission is the test.

Why an ordinary crawler misses all of this

A standard scraping pipeline checks robots.txt, if it checks anything. Robots.txt tells a crawler which paths it may fetch. It does not carry a clause about AI training, a requirement to take a licence or a reference to the Information Technology Act. Those statements sit in terms of use, in copyright and AI policy pages and sometimes in Hindi.

What is read
Illustration. One block is robots.txt; the others are the site's signal files, its terms and policy pages, its licences and a barrier.

A crawler does not open those pages. If it did, it could not tell a refusal from routine legal text. The pipeline therefore goes on collecting from a site whose owner has said no in writing, and nobody on the team knows until a letter arrives.

This is the gap that RightSignal closes. It finds a site's policy pages and reads them in English and in Hindi. It quotes the clause that decides, applies the India profile and seals what the site said and when. With strictest caution on, a clear refusal that invokes Section 43 or Section 66 is treated in the same way as an access barrier. Your team learns of the refusal before it collects, and it holds a dated record of having respected it.

ANI Media v OpenAI, and what it does and does not decide

On 24 July 2026 a single judge of the Delhi High Court refused the news agency ANI an interim injunction against OpenAI. It is the first Indian ruling on training a language model with copyright works, and it is often summarised too broadly.

What the court held, on a first view of the case

  • Storing ANI's freely available news content and using it to train a language model fell, prima facie, within fair dealing under section 52(1)(a) of the Copyright Act, 1957, as private or personal use that includes research.
  • The answers placed before the court were not shown to reproduce ANI's articles in substance. The court considered that they were more likely to have been produced by retrieval at the time of the question than from memorised training material.
  • ANI was able to block OpenAI's crawlers, both for training and for search and retrieval, and had not done so. The court weighed that unused opt-out when it refused the injunction. It did not hold that opting out is a legal requirement.

What the ruling did not decide

  • It did not decide the suit. The court recorded that its observations were made for the interim application only and do not affect the final outcome.
  • It is a copyright ruling. It did not rule on Section 43 of the Information Technology Act or on the effect of a website's terms of use.
  • It is under appeal. ANI's appeal, numbered FAO(OS)(COMM) 237/2026, is before a Division Bench of the same court. The Bench issued notice on 15 September 2026, and the appeal is listed for 8 December 2026.

What a careful business takes from it. The ruling rests on its facts. The content was freely available, no paywall stood in the way and the publisher had not opted out. Where those facts are present, a team can point to the ruling as interim support. Where a site has blocked your crawler, refused in its terms or put up a barrier, those facts are absent and the ruling does not help you. The practical lesson runs against the headline. The ruling makes a site's opt-out matter more, and it makes your record of whether a site had opted out the evidence that counts.

RightSignal's India profile relies on the ruling only as interim authority, and only where the facts the court relied on are present. If the appeal or the trial changes the position, RightSignal will issue a new signed version of the Algorithm.

The DPIIT proposal on training data

In December 2025 a committee set up by the Department for Promotion of Industry and Internal Trade published the first part of a working paper on generative AI and copyright. The paper proposes that AI developers receive a mandatory blanket licence to train on all "lawfully accessed" copyright works. Rightsholders would receive a statutory right to remuneration, collected through a central body at rates set by a committee that the government appoints. Developers would file a disclosure with a sufficiently detailed summary of their training data. The committee did not favour an opt-out model.

This is a proposal and not law. It would need Parliament to amend the Copyright Act, and the RightSignal Algorithm does not apply it. Two points in it are still worth your attention today.

  • The proposed licence would cover only works that were "lawfully accessed", so the access question that Section 43 raises would remain.
  • Developers would have to disclose what they trained on, and they could do that only from records kept at the time.

Whatever becomes of the proposal, a team that knows its sources and can evidence how it reached them is prepared for it.

The DPDP Act and public personal data

The Digital Personal Data Protection Act, 2023 does not apply to personal data that the person concerned has made publicly available, or that another person has made public under a legal obligation. Teams sometimes read this as a general exemption for public web data. It is narrower than that.

A page can be public while the personal data on it was posted by someone other than the person it describes, with no legal duty to publish it. That data stays within the Act, and a crawler cannot tell the two cases apart at scale. The Rules under the Act were notified in November 2025 and take effect in stages. The main duties on those who handle personal data are due to apply from May 2027.

RightSignal reads what a site says about access to its pages and use of its content. It does not assess personal data, and no outcome says anything about the Act. If your sources hold personal data, you need a separate basis for collecting it, and your lawyer should advise you on that.

How RightSignal helps you collect in good faith

  • It reads every place where permission is stated. RightSignal reads robots.txt for your crawler, the site's signal files, its terms and policies and any barrier. You no longer depend on a single file to tell you what the owner wants.
  • It stops where the site says stop. A robots.txt rule against your crawler, a clear written refusal or a barrier produces Do not collect under the India profile. You avoid collecting what you would later have to remove.
  • It quotes and seals. The receipt holds the site's own words, the address they came from and the time they were read. You can show what you knew and when you knew it.
  • It reads Hindi. A refusal in a Hindi terms page is read as a refusal. A page in a language RightSignal does not read is held for review and is never treated as silence.
  • It offers a stricter setting for statutory warnings. Strictest caution treats a refusal that invokes Section 43 or Section 66 as a barrier.
  • It keeps the facts of the interim ruling on record. For each source, the receipt shows whether the site was open, whether a barrier stood in the way and whether the site had refused.

What this lets you show

  • To an enterprise customer

    You can answer a provenance question in a procurement form with receipts that the customer's own team can verify.

  • To an investor

    You can hand over a record for each source in a diligence review, and a Trace report for data collected before you began.

  • To a rightsholder who writes to you

    You can reply with the site's own words and the date you read them, which turns a dispute about facts into a conversation about terms.

  • In a dispute

    Your counsel can show that your team looked for the owner's permission in every place the owner states it and stopped where the answer was no. A sealed record of that conduct is the evidence a lawyer will want when the question is how your team behaved.

If you plan to serve the EU

Many Indian AI companies will want European customers. The EU AI Act applies its copyright duties to any provider that places a general-purpose model on the EU market, wherever the model was trained. If Europe is in your plans, collect under the India and EU Bridge profile from the start. It applies the stricter reading of the two at every step, so the dataset you build for India today carries the records that EU buyers and regulators will ask about.

Collect in good faith, and be able to show it.