Short answer: On 24 July 2026, the Delhi High Court refused ANI’s request for an interim injunction against OpenAI. At that preliminary stage, it held that OpenAI’s temporary storage of ANI material for large-language-model training was prima facie capable of falling within the fair-dealing exception in Section 52(1)(a) of the Copyright Act, 1957. The court also found that ANI’s cited ChatGPT answers did not sufficiently establish memorisation or substantial reproduction.
That is not a ruling that every form of commercial AI training, scraping or retrieval is lawful. The judgment shows that India’s existing statute can produce an answer in a particular dispute, but it does not provide a complete text-and-data-mining regime for modern AI.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Copyright Law | $144.88 | Buy on Amazon |
| 2 |
|
Copyright Law: Cases and Materials (v8.0) | $19.87 | Buy on Amazon |
| 3 |
|
Copyright Law of the United States: and Related Laws Contained in Title 17 of the United States Code | $10.32 | Buy on Amazon |
| 4 |
|
Copyright Law in a Nutshell | $65.00 | Buy on Amazon |
| 5 |
|
Copyright Handbook, The: What Every Writer Needs to Know | $37.99 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
What the Delhi High Court actually decided
In ANI Media Pvt. Ltd. v. Open AI OpCo LLC, Justice Amit Bansal dealt with ANI’s application for interim relief. The judgment is available through the reported judgment; the Delhi High Court’s official site is delhihighcourt.nic.in.
ANI presented two related but legally different complaints:
#1 Best Overall
- Training: OpenAI allegedly copied and stored ANI’s literary works while preparing and training the models behind ChatGPT.
- Outputs: ChatGPT allegedly generated answers that reproduced or substantially copied ANI material.
The court rejected the interim-injunction request. Its conclusions were preliminary, based on the evidence and examples before it, and do not replace a final trial judgment or any later appellate decision.
One lawsuit, several technically different acts
“AI copyright infringement” is not a single event. A system may perform several acts, each raising a different legal question.
| Technical stage | Possible legal issue |
|---|---|
| Web crawling or downloading | Making a copy of a protected work |
| Dataset preparation and temporary storage | Electronic reproduction, even if the copy is later deleted |
| Tokenisation and other processing | Whether a transformed representation remains legally relevant |
| Training and model-weight formation | How reproduction applies when the source text is not delivered as a document |
| Retrieval at answer time | Potentially direct delivery of source material |
| Generated output | Whether protected expression, rather than merely facts, has been substantially reproduced |
| Publication or commercial exploitation | Economic harm, market substitution and available remedies |
A conclusion about temporary training copies cannot automatically be transferred to a retrieval-augmented system that stores and quotes articles at answer time, or to a chatbot that emits a distinctive passage verbatim.
Why servers outside India did not end the case
OpenAI argued that relevant servers were located outside the United States. The court nevertheless held that Indian courts could entertain the dispute and separately considered how Indian copyright law applied to the claims.
Three questions must be kept separate:
- Jurisdiction: whether an Indian court can hear the case.
- Applicable law: which country’s substantive copyright rules govern the conduct.
- Enforcement: whether an order can practically affect systems, entities or servers abroad.
Indian users, commercial activity and effects on Indian publishers can matter to jurisdiction. Jurisdiction, however, did not mean that ANI had proved infringement.
Rank #2
How Sections 14, 51 and 52 structure the dispute
Section 14: the exclusive right
Section 14 of the Copyright Act gives an owner exclusive rights in a work, including reproduction. For a literary work, reproduction includes storing it electronically. The statutory text is available on India Code.
That matters because an AI developer may create several copies during ingestion, cleaning and training, even when the final model does not contain a readable copy of every article.
Section 51: when infringement occurs
Section 51 supplies the infringement framework. The analysis generally asks whether a Section 14 right was implicated, whether the defendant performed a reserved act, whether a Section 52 exception applies, and whether the claimant owns protected material.
Section 52(1)(a): the exception at issue
Section 52 contains exceptions to infringement. The relevant fair-dealing provision refers to purposes including private or personal use and research. The Copyright Office publishes the exceptions at copyright.gov.in.
The court’s reported approach separated two questions:
Rank #3
- Purpose: does the use fall within a purpose named in Section 52(1)(a)?
- Fairness: even if the purpose is covered, is the dealing fair on the facts?
The judgment treated OpenAI’s temporary storage for model training as prima facie capable of protection under this provision. It did not create a blanket exemption for commercial AI companies. Commerciality can still affect fairness, economic harm and market substitution.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Why the ruling does not make all AI training legal
India has no provision expressly labelled a generative-AI training or text-and-data-mining exception. A Department for Promotion of Industry and Internal Trade working paper identified that gap and questioned how far Section 52(1)(a) can reach; the paper is not legislation. See the DPIIT working paper.
Existing language can be interpreted for new technology, but interpretation is not the same as a predictable industry rule. Future cases could involve books, music, photographs, software, databases, paywalled material, pirated datasets, confidential information or personal data. The facts, ownership, contracts, originality and market effects may differ substantially.
What the court said about ChatGPT outputs
ANI’s output claim depended on particular prompts and answers. The court found insufficient evidence in the cited examples to establish that ChatGPT had memorised and substantially reproduced ANI’s works. The analysis considered works as a whole rather than treating isolated fragments as decisive.
Facts are not the same as expression
A chatbot can report the same event or fact without copying ANI’s wording. Copyright protection is generally concerned with original expression, including distinctive phrasing, arrangement and selection. Repeatedly producing a long, unusual passage, sequence or quotation would present a stronger case than merely giving the same factual account.
Free tools Windows power users keep installed
One-click scans. No signup required.
Rank #4
Memorisation needs evidence
Useful evidence could include repeated verbatim outputs, long distinctive passages, prompt sensitivity, comparisons with the complete source, and technical testing that distinguishes memorisation from independent generation. Similar facts alone do not establish that a model retained and reproduced an article.
The competing public interests
The court’s reasoning reflects a conflict that cannot be reduced to “ANI versus OpenAI.”
Rights-holder concerns
- Electronic storage can engage the reproduction right.
- Publishers fund reporting while AI services may monetise access to the resulting information.
- Unlicensed training could weaken licensing markets and reduce incentives for original journalism.
- Limited dataset transparency makes it difficult to prove that a particular work was used.
AI-industry concerns
- Training is argued to be transformative and ordinarily does not deliver a copy of every source article.
- Licensing every item in a very large corpus may be administratively and commercially difficult.
- Outputs should be assessed for substantial similarity, not simply factual overlap.
- A broad injunction could affect Indian users and domestic AI development.
The public-interest balance
The court also considered access to AI tools, technological development, competition and public dissemination of knowledge alongside protection for journalism. That balancing may support an interim decision, but it does not answer how creators should be compensated or how a durable licensing market should work.
What the judgment leaves unresolved
- Whether model training itself is always reproduction under Section 14.
- Whether temporary copies, tokenised data and model weights should receive the same treatment.
- How commercial purpose should affect fair dealing in different datasets.
- What transparency or disclosure duties AI companies should have.
- Whether creators should receive remuneration, an opt-out right or collective licensing.
- How courts should handle outputs generated through retrieval rather than training alone.
- Which remedies are proportionate if infringement is ultimately proved.
Possible remedies could include injunctions, damages, licensing, disclosure, deletion or retraining, output controls, attribution or negotiated collective remuneration. An injunction is not the only possible response.
Common misreadings of the ANI decision
“The court said AI training is legal in India”
Too broad. The decision was interim and prima facie, concerning ANI’s evidence and pleaded claims.
Best Value
“Copyright does not cover AI training”
Misleading. The court recognised that storage could engage reproduction rights before considering the Section 52 exception.
“OpenAI definitively won”
OpenAI prevailed on the interim-injunction application, not necessarily on every final merits issue.
“ANI lost because its content was public”
Public availability is not a waiver of copyright. It was only part of the factual and fairness context.
Recommended Free Tools
“ChatGPT cannot infringe copyright”
The court found the cited outputs insufficient on the evidence before it. A different answer could produce a different result.
What creators and publishers should do now
- Keep publication records, dated copies and ownership or assignment documents.
- Monitor suspiciously verbatim or unusually long chatbot outputs.
- Save the exact prompt, response, account, date and URL when preserving evidence.
- Separate factual overlap from copied wording, structure or quotations.
- Review crawler policies, licences and contractual restrictions.
- Use technical access controls where appropriate, without assuming they settle the legal question.
- Obtain Indian copyright advice before sending a takedown demand or filing suit.
What AI companies should do now
- Maintain provenance records for training and evaluation data.
- Separate licensed, public, user-provided and uncertain sources.
- Document acquisition methods, retention periods and deletion practices.
- Test for memorisation and long-form reproduction.
- Implement output safeguards and a complaint and correction channel.
- Preserve relevant logs and dataset evidence for possible litigation.
- Analyse Indian law separately from United States or European rules.
Practical sources and professional help
For primary materials, start with India Code, the Delhi High Court and the Copyright Office of India. Commercial research services such as Westlaw and Lexis+ may offer broader citator and case-search functions, but coverage, eligibility and pricing vary.
An enterprise AI subscription, including information at OpenAI’s business site, does not itself grant permission to ingest a publisher’s copyrighted corpus. Companies handling training data or cross-border disputes should consult Indian copyright counsel.
Bottom line
India’s Copyright Act, 1957 was flexible enough for the Delhi High Court to deny ANI’s interim application. The court separated training copies from generated outputs, accepted that temporary storage could engage reproduction rights, and found a possible fair-dealing route under Section 52(1)(a).
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →But the statute still lacks detailed rules for industrial-scale AI training, dataset transparency, opt-outs, remuneration, model memorisation and retrieval systems. The ANI decision is therefore an important interpretive step—not a comprehensive licence to scrape, train or reproduce copyrighted material.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
