Delhi High Court’s ANI v. OpenAI Ruling: India’s First AI Training Copyright Verdict Rewrites Fair Dealing for the LLM Era

A Watershed Moment for Indian Copyright Law

The Delhi High Court has delivered what practitioners are calling India’s most consequential copyright ruling in a generation. In ANI Media Pvt. Ltd. v. OpenAI Inc., the Court held — prima facie — that OpenAI’s storage of ANI’s literary works for training the large language models underlying ChatGPT falls within the protective ambit of Section 52(1)(a) of the Copyright Act, 1957, and does not amount to infringement. This is the first time an Indian court has squarely addressed whether AI training on copyrighted works constitutes fair dealing, and the answer will reverberate through every IP practice in the country.

The Court’s Reasoning: Four Pillars

The ruling rests on a carefully constructed edifice. First, the Court characterised the use as “private or personal use, including research” under Section 52(1)(a) — a reading that stretches the conventional understanding of “private use” to encompass the computational ingestion of millions of works by a commercial entity. Second, the use was found to be transformative — the works were used solely for training purposes, not for reproduction or distribution in their original form. Third, the Court emphasised that the training did not substitute the market for ANI’s original works. Fourth, and perhaps most significantly, the Court invoked the public interest in innovation and knowledge dissemination as a factor weighing against a finding of infringement.

This is not merely a copyright case. It is a statement about how India intends to position itself in the global AI race — and the judiciary has signalled that it will not let legacy copyright doctrine strangle technological development.

Why This Matters for Practitioners

The immediate implications are profound. Content owners — news organisations, publishers, stock photo agencies, music labels — have just lost their first major battle against AI developers in India. The ruling effectively creates a safe harbour for AI training under the existing fair dealing framework, without any legislative amendment. For practitioners advising AI companies, this is a green light. For those advising content creators, it is a wake-up call to pursue legislative remedies or pivot to contractual protections and technological measures.

But the ruling’s reasoning deserves closer scrutiny. The characterisation of commercial AI training as “private or personal use” is, frankly, a stretch. Section 52(1)(a) was drafted in an era when “private use” meant a researcher photocopying a chapter in a library. To extend it to the industrial-scale ingestion of copyrighted content by a company valued in the tens of billions requires considerable judicial creativity. The Court’s reliance on the transformative nature of the use and the absence of market substitution echoes the American fair use doctrine more than traditional Indian fair dealing analysis, which has historically been narrower and more category-bound.

The Prima Facie Caveat

It is critical to note that this is a prima facie finding — likely at the interim relief stage. The matter will proceed to trial, and the final outcome may differ. However, as any seasoned IP litigator knows, interim findings in copyright cases often harden into final positions. The persuasive weight of this ruling, particularly given its first-of-its-kind status, will be enormous.

The Broader Context: DPDP, AI Authorship, and Legislative Gaps

This ruling does not exist in a vacuum. It arrives amid a flurry of developments at the intersection of AI, data, and IP law in India. The DPDP Act’s phased implementation — with substantive obligations for Data Fiduciaries taking effect on 13 May 2027 — is creating a parallel regulatory framework that will govern how personal data is used in AI training. The Removal of Difficulties Order, 2026, notified by MeitY on 5 October 2026, has already begun clarifying consent requirements and data audit obligations, replacing the word “audit” with “data audit” in Section 10(2)(c)(ii) — a seemingly minor change that signals the government’s intent to impose meaningful accountability on data processors.

Meanwhile, the question of AI authorship remains unresolved. As Bar and Bench has noted, the emerging position is that AI-generated works can attract copyright protection, but the AI system itself cannot be named as an author. This aligns with the position under Section 2(d)(vi) of the Copyright Act, which recognises the “person who causes the work to be created” as the author in the case of computer-generated works — but the contours of this provision in the generative AI context remain untested.

What Comes Next

Three developments bear watching. First, the final judgment in ANI v. OpenAI — if the prima facie view is upheld, it will become the definitive authority on AI training and copyright in India. Second, the Supreme Court’s larger bench decision on the DPDP Act’s constitutional validity, particularly the challenge to Section 44(3) and its amendment of the RTI Act, which could reshape the data landscape in which AI companies operate. Third, the possibility of legislative intervention — either through amendments to the Copyright Act to create a specific text-and-data-mining exception (as Japan and the EU have done) or through the proposed trade secrets legislation that is reportedly heading for a fresh draft.

For now, the Delhi High Court has planted its flag firmly on the side of innovation. Whether Parliament or the Supreme Court will ratify that position remains the defining question for India’s AI ecosystem.