No news content for AI-training without licences

LLM developers would be well advised not to violate copyright protections based on misinterpretations of the Delhi HC's interim order in ANI vs. OpenAI.

The Delhi High Court’s interim order in ANI Media Pvt. Ltd. v. OpenAI OPCO LLC (2026) refusing injunctive relief, may be a precursor to recasting India’s copyright laws and interpreting existing laws in light of evolving technologies, to ensure transparency and effective balancing of owner rights with that of public interest. Further appeals and developments on this order are inevitable.

A final decision in ANI vs. OpenAI would go a long way in laying down jurisprudence on the intersection of AI and copyrights. (Stock Adobe)

This, being an interim order, does not decide rights of all large language models (LLMs), but its impact in evaluating the legality of usage of copyright-protected content in training and reproductions by LLMs cannot be disputed. This being an interim order, it would be premature for LLM developers to rejoice or assume unfettered usage of publicly available data. For the public, sans technicalities, the relevance lies in continued availability of technology-enabled conveniences. For the copyright owners, this is a wake-up call to strengthen their rights through statutory upgrades and use of technology to protect their publicly shared proprietary content.

A copyright owner’s rights over proprietary content restrict third parties from reproducing and communicating to public such content without consent of the owners. Section 14 of the Copyright Act, 1957 (the Act) explicitly restrains “storing” of such data without owner consent. The question of LLMs being trained using copyright protected content scraped from the web, without consent of the owner of such data, cuts to the root of protections afforded under the Act.

In ANI v. OpenAI, the Delhi High Court delves into copyright and LLM fundamentals, and upholds ANI’s copyright claims to its news content, along with the reiteration that “original literary works” that are available free of cost and publicly accessible are nevertheless entitled to copyright protection. This is critical to combat rampant plagiarism.

However, where the Delhi High Court based its refusal of injunctive relief is the factual comparison of ANI’s claims of violation with actual reproductions by ChatGPT. The HC found the commonalities to be inadequate. The issue of whether LLMs memorise and regurgitate to an extent that can be considered infringements of proprietary content is left open to a decision in trial. ANI’s failure to demonstrably show substantial reproductions that warrant restrictions on OpenAI is a factual finding and will not impact cases where copyright owners are able to show substantial infringement. If the reproductions are identical or substantially similar to the copyrighted content, injunctive reliefs will follow, as was the outcome in GEMA v. OpenAI, decided in the Regional Court of Munich.

The remaining issues framed and decided at a preliminary stage are the legality of storing proprietary data and “fair use” exemptions. The findings here are most likely to form the basis for an appeal by ANI and any other copyright owner to distinguish from this interim order. Trends across jurisdictions point to both restrictions on misuse of publicly available data, as in the GEMA case, and permissibility under fair dealing, including the three-factor test formulated by the HC. Critically, while extensive submissions were made interpreting “derivative” and “transformative” works, the same does not form the basis for the final conclusion of the Court.

Copyright protection under the Act is ringfenced by exemptions under Section 52. Under Section 52(1)(a), “fair dealing of any work”, among other things, for “private or personal use, including research” would be non-infringing. The Explanation further qualifies that “storing” in any medium for the purposes in this provision would not amount to infringement. Section 14, which restricts storing by third parties without licence from owners, has been read with Section 52(1)(a) by the HC to hold that while, in general, “storage” of copyright-protected data is an infringement, it would not be so when it falls within the exemptions. The HC follows the Supreme Court of Canada in CCH Canadian v. Law Society of Upper Canada, while liberally interpreting the “private or personal use including for research”. The court does emphasise that the “research is at the behest of humans and for the benefit of humans”, but then fails to clarify how training of non-human technology would qualify under this exemption.

On “fair dealing”, the Court formulates a three-factor test of whether OpenAI used ANI’s data only for training its LLMs, whether the outcome causes actual or potential damage to the business of ANI, and whether the functions performed by ChatGPT serve public interest. On these, the Court holds that ChatGPT’s outputs are distinct from those of ANI and that extracts cannot harm the business prospects of ANI. With respect to the first test, the Court relies on factual findings to hold that Open AI has only used the data for training. The Court’s stand with respect to the third test remains tenuous as convenience has been confused with public interest. The need to clarify primacy of ownership rights over convenience, therefore, still remains to be tested by judicial review. Public interest lies in having access to copyright-protected content and not in infringement thereof. Such access can be assured even through licensing, including compulsory licensing in case of hoarding of knowledge.

It remains to be seen which among the many grounds for rejection of injunctive reliefs will sustain further judicial review. A final decision in this case would go a long way in laying down jurisprudence on the intersection of AI and copyrights. Meanwhile, developers would be well advised to proceed with caution and to not violate copyright protections on misconceived interpretations of the HC judgment. Copyright owners need to tighten their technology enabled protections, while battling for clarity in the statutes.

NS Nappinai is senior advocate, Supreme Court, and founder, Cyber Saathi. The views expressed are personal.