Delhi High Court Examines Copyright and AI Training

The Delhi High Court's interim judgment fundamentally misreads Section 52 of the Copyright Act by extending "research" protection to commercial-scale ingestion of proprietary news content by a for-profit entity. ANI's position is straightforward: fair dealing under Section 52(1)(a) was crafted for human scholarship, criticism, and review — not for the automated, wholesale copying of an entire news archive to build a competing commercial product that substitutes for the original. The court's invocation of "technological neutrality" cannot override the statutory text, which requires the dealing to be "fair" and for specified purposes; a machine training run that reproduces expressive content verbatim fails both limbs. OpenAI's own contention that its activities qualify as research does not make it so — the statute protects the purpose, not the actor, and the purpose here is commercial model-building, not research in any recognised sense. The judgment creates a precedent that would allow any commercial actor to ingest copyrighted works at scale without licence, destroying the economic basis of original news gathering.

ANI argues that Section 52(1)(a) protects only human scholarship and that commercial model-building falls outside "research" because the purpose is profit-driven. The Delhi High Court rejected this binary, holding that technological neutrality requires the law to focus on the nature of the activity — not the identity of the actor — and found OpenAI's training process prima facie constitutes research protected under fair dealing (Indian Express). The court's 135-page judgment expressly observed that machines now perform tasks once the sole remit of humans, and that the statutory purpose of research is served by the analytical extraction of patterns from data, irrespective of whether a human or system performs it (Indian Express). ANI's framing conflates the commercial status of the entity with the character of the use; the Copyright Act does not strip fair dealing protection merely because the researcher is a for-profit company.

OpenAI correctly paraphrases the court's technological-neutrality reasoning but sidesteps the statutory text: Section 52(1)(a) protects "fair dealing… for the purposes of private use, research, criticism or review" — a purpose-defined, not actor-defined, list. The court's observation that machines now perform human tasks (Indian Express) does not answer whether the dealing itself is "fair" when it ingests an entire proprietary archive to build a commercial substitute that displaces the original market. Fairness under Indian law turns on the quantum taken, the purpose, and the effect on the potential market; a training run that copies expressive content wholesale and enables verbatim reproduction fails all three limbs regardless of whether a human or a model performs the extraction. The judgment's prima facie finding collapses the distinction between analytical pattern extraction and expressive copying — precisely the line Section 52 was drawn to protect.

ANI's rebuttal asserts that "quantum taken" and "market displacement" defeat fair dealing regardless of technological neutrality, but the court's prima facie finding necessarily engaged those very factors — it concluded the analytical extraction of patterns from ANI's works, not expressive copying, constitutes protected research (Indian Express). The judgment's 135-page reasoning distinguishes between ingesting data to learn statistical relationships and reproducing expression for consumption; ANI conflates the two to argue that any large-scale ingestion is per se unfair. That conflation would make the Copyright Act hostile to any computational research at scale, a result the court expressly rejected by affirming that the statute's purpose is served by pattern extraction irrespective of the actor (Indian Express). The interim ruling does not collapse the expressive/analytical line — it applies it, and finds OpenAI's training on the analytical side.