Does training an AI system on someone else’s copyrighted content amount to infringement? In ANI Media Pvt. Ltd. v. Open AI OpCo LLC, the Delhi High Court refused an interim injunction and held that storing content for LLM training falls within the fair dealing exception under Section 52 of the Copyright Act, 1957. Barely weeks earlier, in Keshan Infotech Pvt. Ltd. v. Oliver Brandt & Ors., the Madras High Court reached what looks like the opposite conclusion at the interim stage, holding that unauthorised use of proprietary content as AI or LLM training data raises a prima facie case of infringement serious enough to justify immediate injunctive relief. Read together rather than in isolation, the two orders say less about a doctrinal split and more about how much weight the specific facts, the procedural posture and the identity of the parties carry in this still developing area.
What Keshan Infotech actually alleged
New to Copyright Law? Start with our complete Copyright Law Guide.
Keshan Infotech Pvt Ltd, a Kolkata based company operating the travel website travelandtourworld.com, filed a commercial suit against an individual named Oliver Brandt, an editor based in Verona, Italy, along with Google LLC, Google India Private Limited and an unidentified John Doe respondent described as Ashok Kumar. The applicant’s case, built around original travel articles it said it had produced as a travel vlogger and content creator, was that Brandt had been systematically scraping its published material and republishing it on Google owned social media platforms. The applicant pointed to a specific circumstance that made the allegation harder to dismiss as speculative, its own logo appeared to have been retained on the republished content even as the original authorship attribution was stripped out or masked. A social media post from Brandt himself, annexed to the plaint, was cited by the court as lending prima facie support to these claims.
Keshan Infotech sought four separate ad interim injunctions through applications numbered OA 637 to 640 of 2026, filed within the larger commercial suit registered as C.S(COMM DIV) No. 181 of 2026. The relief sought was drafted with unusual specificity for the AI era. The first injunction covered reproduction, copying, translation, adaptation, summarisation or paraphrasing of the applicant’s content, whether carried out by a human being or through any AI assisted or algorithmic process. The second went further and asked the court to restrain use of the applicant’s content as an input, source, reference, training data or prompt for any AI model, large language model, automated content generation tool or web scraper, for any purpose whatsoever. The third sought to prevent any automated scraping tool, AI system or content generation mechanism from being pointed at the applicant’s website at all, whether directly or indirectly. The fourth asked for protection against the applicant’s work being reproduced in any transformed, paraphrased, summarised or derivative form, again including through AI or automated means.
The order and what the court actually decided
Hon’ble Justice K. Kumaresh Babu granted all four applications, allowing an ad interim injunction for a period of four weeks and directing that notice be issued to the respondents, returnable within the same four week window, with permission for the applicant to effect private service through speed post given that one respondent was based abroad. The order records the court’s view that a prima facie reading of the plaint, supported by the annexed social media material, disclosed a case of copyright infringement warranting immediate intervention, and that this extended specifically to the unauthorised use of proprietary content as training data or prompts for AI systems.
It is worth being precise about what this order does and does not settle. It is an ad interim order, granted at the very first hearing, on the applicant’s version of events alone, without the respondents having had an opportunity to appear and contest the claims. The four week timeline built into the order itself reflects this, the injunction is explicitly temporary and the matter is due to come back before the court once notice has been served and the respondents have had a chance to respond. Nothing in the order forecloses a different outcome once Google, as an intermediary respondent, or Brandt, as the alleged primary infringer, actually contest the application. What the order does establish, at least for now, is that an Indian court was prepared to treat the prospect of AI training on scraped content, dressed up with a preserved logo and a stripped byline, as serious enough to warrant relief without waiting for a contested hearing.
Why this sits uneasily next to the ANI ruling
The comparison to ANI Media Pvt. Ltd. v. Open AI OpCo LLC (Read full case analysis here: ANI v. Open AI Analysis) is instructive precisely because the two cases are not, on their facts, doing the same thing. Justice Amit Bansal’s judgment in the ANI matter came after a fully contested interim injunction hearing spanning many months, involving detailed evidence, expert opinion and submissions from senior counsel on both sides, and it addressed the question of whether OpenAI’s storage of ANI’s publicly available content for training ChatGPT fell within Section 52(1)(a) as research use. The Delhi High Court found that it did, reasoning in part that developing a large language model would become economically unworkable if a developer needed to license every source individually, and that ANI’s own failure to use available technical measures to block web crawlers from its site weighed against it.
Keshan Infotech’s case, by contrast, was decided at the very first, ex parte stage of the proceedings, on an entirely different fact pattern involving an identifiable individual accused of directly republishing scraped articles under a masked byline while keeping the original publisher’s logo intact, conduct that reads less like automated large scale model training and more like straightforward content misappropriation with an AI flavoured pleading layered on top. The prayer clauses themselves, covering both direct republication and AI training use in the same breath, suggest the applicant was hedging against multiple possible uses of its scraped content rather than building a case squarely about LLM training in the way ANI’s suit against OpenAI does.
Even accounting for those differences, the tension is real and will not resolve itself simply by pointing to procedural posture. If unauthorised scraping for AI training purposes is capable of supporting an ad interim injunction in one court on a prima facie reading of a plaint, while a fully contested hearing in another court finds the same underlying activity protected as fair dealing, litigants and platforms operating across India are left without a consistent signal on how to structure their data collection and AI training practices. Much will depend on what happens when Google and Brandt actually appear in the Keshan Infotech matter and place their own evidence before the court, and equally on what the Delhi High Court eventually decides in the still pending main suit in the ANI litigation, since the July ruling there was itself only at the interim stage.
The Google question sitting quietly in the background
One detail in the pleadings deserves more attention than it has received so far. Keshan Infotech named both Google LLC and Google India Private Limited as respondents alongside Brandt, on the basis that the republished content was appearing on Google owned social media platforms. This puts an intermediary directly inside an AI training dispute rather than keeping the fight confined to the alleged scraper and the original publisher, and it raises questions the current four week order does not begin to answer. Safe harbour protection under Section 79 of the Information Technology Act ordinarily shields an intermediary that merely hosts third party content and acts on takedown notices, but that protection has its own conditions attached, including a requirement that the intermediary not have actual knowledge of the infringing material and that it act with reasonable diligence once notified. Whether a platform that hosts republished, logo intact but byline stripped travel content, and whose own infrastructure or associated tools might feed that content into downstream AI systems, can comfortably rely on that shield is a separate question the Madras High Court has not yet been asked to decide, since the current order runs only against the named individual respondent pending further hearing. If Google does appear and contest its inclusion in the suit, the arguments it raises will likely track the same safe harbour jurisprudence that has shaped earlier Indian intermediary liability disputes, and the outcome could end up mattering as much to platforms hosting AI adjacent content as the training question itself does to model developers.
Also Read: Disney, Universal & Warner Bros v. MiniMax: AI Copyright
Conclusion
It would be a mistake to read either of these orders as the final word on how Indian copyright law treats AI training, and I say that as someone who expects this exact question to occupy a meaningful share of IP litigation over the next several years. What strikes me most about the Madras order is how quickly the court was willing to extend interim protection against AI training uses specifically, rather than treating that prayer as an afterthought tacked onto a more conventional republication claim, and I suspect that willingness reflects a broader judicial instinct that content creators deserve some measure of protection while the larger legal questions get worked out through fuller hearings elsewhere. At the same time, an ex parte order obtained on one side’s pleadings carries far less precedential weight than a reasoned judgment delivered after contest, and treating the two rulings as directly contradictory overstates what either court has actually decided so far. The honest position, for now, is that Indian courts are approaching AI training disputes case by case, weighing the specific conduct alleged, the identity and resources of the respondent, and the stage of the proceeding, and that a coherent doctrinal position on how Section 52 applies to AI training will only emerge once matters like these are argued out fully rather than decided on an interim, one sided basis.
Case: Keshan Infotech Pvt. Ltd. v. Oliver Brandt & Ors., OA Nos. 637 to 640 of 2026 in C.S(COMM DIV) No. 181 of 2026, Madras High Court, ad interim order dated 1st July 2026.
