OpenAI vs ANI case: what it means for the future of information
The Delhi High Court declined to grant an interim injunction sought by news agency ANI against OpenAI, holding that using ANI's content to train a large language model is, on a prima facie basis, covered by "fair dealing" under the Copyright Act, 1957.
The order addresses only ANI's application for interim relief; the main copyright infringement suit, filed by ANI in 2024, remains pending for full trial.
The court's reasoning centred on Section 52(1)(a) of the Copyright Act, which exempts fair dealing with a work for research, private/personal use, criticism, review, or reporting of current events from constituting infringement.
The order is described as among the first detailed Indian judicial engagements with how large language model training fits within existing domestic copyright law.
Section 52(1)(a), Copyright Act, 1957 — Fair Dealing Exception
Section 52 of the Copyright Act, 1957 lists specific acts that do not constitute copyright infringement. Section 52(1)(a) exempts fair dealing with a literary, dramatic, musical, or artistic work for the purposes of (i) private or personal use, including research, (ii) criticism or review, and (iii) reporting of current events, from infringement liability.
Key Details
- The Copyright Act, 1957 is India's principal copyright statute, substantially amended in 1994 and 2012 (the latter primarily to address digital-era and disability-access issues).
- India's "fair dealing" doctrine is purpose-bound and enumerated — unlike the open-ended, four-factor US "fair use" test — meaning a use must be shown to fall within one of the listed statutory purposes (here, "research").
- The court held that storing and processing ANI's content to train OpenAI's models fell within the "private or personal use, including research" limb of Section 52(1)(a)(i), at least at the interim stage.
The ruling's core legal basis is this specific statutory limb — training data ingestion was characterised as a form of "research," bringing it within the fair-dealing exception rather than treating it as reproduction/distribution of the copyrighted work.
Comparative Context: AI Training and Copyright Litigation Abroad
Parallel disputes over AI training data have been contested in other jurisdictions under different legal standards, most notably the US "fair use" doctrine under 17 U.S.C. Section 107.
Key Details
- The New York Times' suit against OpenAI and Microsoft (US, ongoing) and rulings in Bartz v. Anthropic and Kadrey v. Meta have examined whether training generative AI models on copyrighted text is a "transformative" fair use.
- The US four-factor fair use test weighs: purpose/character of use, nature of the copyrighted work, amount and substantiality used, and effect on the market for the original work — a broader, judicially balanced standard than India's enumerated fair-dealing purposes.
- India currently has no explicit statutory text-and-data-mining (TDM) exception, unlike the European Union's 2019 Copyright Directive (Articles 3 and 4), which carves out specific TDM exceptions for research and general purposes.
Because Indian law lacks a dedicated TDM exception, the Delhi High Court had to fit AI training within the existing "research" limb of Section 52(1)(a) rather than apply a bespoke or open-ended standard, distinguishing the Indian approach from both the US and EU frameworks.
Interim Injunction Standard in Indian Civil Procedure
Grant of a temporary injunction under Indian law requires the applicant to establish three elements: a prima facie case, balance of convenience in the applicant's favour, and irreparable injury if the injunction is refused — a standard settled by the Supreme Court in Wander Ltd. v. Antox India Pvt. Ltd. (1990) and applied through Order XXXIX, Rules 1-2 of the Code of Civil Procedure, 1908.
Key Details
- The Delhi High Court's finding that ANI did not establish a strong prima facie case of infringement (given the applicability of Section 52(1)(a)) was sufficient, on its own, to justify refusing the interim injunction.
- Refusal of interim relief is distinct from a final adjudication on infringement — the underlying suit continues to trial, where the fair-dealing defence could still be tested on a fuller evidentiary record.
This distinction is a common source of confusion in reporting on the order — the ruling is a preliminary, interim-stage finding, not a final determination that OpenAI's use of ANI's content is lawful.
Input-Stage Training vs. Output-Stage Reproduction in AI Copyright Analysis
Copyright analysis of generative AI systems typically distinguishes between the "input stage" (copying/ingesting data to train a model) and the "output stage" (the model's generated output potentially reproducing protected expression); courts across jurisdictions have generally shown more openness to input-stage training as a permissible use when the process is statistical/pattern-learning rather than verbatim reproduction or distribution.
Key Details
- India currently has no standalone legislation specific to AI and copyright; the existing 1957 Act (as amended) governs by extension.
- The Delhi High Court's interim finding concerned only the input/training stage; questions about outputs that might reproduce ANI's specific content remain part of the pending main suit.
The order effectively treats input-stage training as protected activity under the existing fair-dealing framework at the interim stage, leaving output-related and broader policy questions for the full trial or eventual legislative reform.
- ANI filed its copyright infringement suit against OpenAI: 2024 (the first Indian news organisation to do so)
- Delhi High Court's interim order refusing injunction: delivered July 24, 2026
- Statutory basis for the fair-dealing finding: Section 52(1)(a)(i), Copyright Act, 1957 ("private or personal use, including research")
- Comparable ongoing foreign litigation: New York Times v. OpenAI/Microsoft; Bartz v. Anthropic; Kadrey v. Meta
- India's fair dealing test: purpose-bound/enumerated (research, private study, criticism, review, reporting current events) — narrower in structure than the US's open-ended four-factor fair use test
- ANI's main copyright suit against OpenAI: remains pending for full trial