The Lift Line
An interim order that says “no injunction” is not the same as an order that says “no infringement.”
Why This Editorial Matters for Your Exam
Answers on AI regulation in India routinely cite the ANI vs OpenAI case as though the Delhi High Court had settled whether AI training on copyrighted content is legal. It has not. This editorial supplies the sharper distinction a strong answer needs: an interim order refusing injunctive relief on a prima facie reading of fair dealing is a procedural finding for a specific application, not a final ruling on the merits, and the gap in Indian copyright law that makes the underlying question genuinely open, the absence of a text and data mining exception, is the real analytical point.
GS Paper 2: Government policies and interventions; issues arising from the design and implementation of policy; statutory bodies and their role.
GS Paper 3: Science and technology, developments and applications in everyday life; issues relating to intellectual property rights (IPR).
| Concept | Meaning | Why it is testable |
|---|---|---|
| Fair dealing (Section 52) | Enumerated statutory exceptions to copyright infringement: private/research use, criticism/review, current-events reporting | The exact exception the Delhi HC invoked, prima facie, for AI training |
| Text and data mining (TDM) exception | A dedicated statutory carve-out permitting automated extraction from copyrighted works, present in EU law (Article 4, DSM Directive) but absent in India | The structural gap the editorial identifies |
| Prima facie finding | A provisional finding made for the limited purpose of deciding interim relief, not a final ruling on the merits | The distinction the editorial says is being misread |
| Interim injunction | A temporary court order restraining an action pending final trial; here, ANI’s application for one was denied | Denial of injunction does not equal a finding of lawfulness |
| DPIIT licensing proposal | December 2025 government proposal for a mandatory royalty framework for AI training on copyrighted content | Shows the executive treats the copyright gap as unresolved, in parallel with the litigation |
Background and Context
ANI Media Pvt Ltd filed suit against OpenAI Inc in the Delhi High Court in November 2024, alleging that ChatGPT was trained on its copyrighted news content without authorisation and that the model could reproduce or closely paraphrase ANI’s reporting in response to user prompts.
On 24 July 2026, Justice Amit Bansal denied ANI’s application for an interim injunction. The order held, on a prima facie basis, that training a large language model on copyrighted content could fall within the fair dealing exception under Section 52(1)(a) of the Copyright Act, 1957; that ChatGPT’s Retrieval Augmented Generation (RAG) architecture produced outputs that were not substantially similar to ANI’s original reporting; and it rejected OpenAI’s objection that the Delhi High Court lacked territorial jurisdiction. The Court was explicit that these were findings for the limited purpose of the interim application, and the suit itself remains pending toward a full trial.
The Digital News Publishers Association (DNPA), whose members include large Indian publishers, subsequently sought to join the proceedings, indicating that Indian news organisations more broadly see the outcome as consequential for their industry, not only for ANI.
The Analysis
1. An interim order and a final ruling answer different questions. The Delhi High Court decided whether ANI had shown enough, at this early stage, to justify restraining OpenAI before trial. It did not decide, finally, whether OpenAI’s use of ANI’s content constitutes infringement. Treating the denial of injunction as equivalent to a declaration of lawfulness collapses two distinct legal questions into one, which is the exact misreading the editorial warns against.
2. Section 52’s exceptions were not built for this scale of use. Private or personal use, criticism or review, and current-events reporting are narrow, enumerated categories from a 1957 statute. Applying them to the automated, wholesale ingestion of a news agency’s archive by a commercial AI model is a significant interpretive extension, one a trial court may revisit with fuller evidence and argument than an interim hearing allows.
3. India has no purpose-built exception for this exact activity. The European Union’s Digital Single Market Directive gives text and data mining a dedicated exception under Article 4, permitting reproduction and extraction from lawfully accessed works, including for commercial purposes, unless a rights holder opts out through machine-readable means. Indian law has no equivalent; fair dealing under Section 52 is being asked to do work it was not drafted for, which is precisely why the question remains unsettled rather than answered.
4. The government’s own parallel move undercuts a settled-law reading. The Department for Promotion of Industry and Internal Trade (DPIIT) proposed, in December 2025, a mandatory licensing and royalty framework for AI training on copyrighted content, including a proposed collecting body to route payments to rights holders. A ministry drafting a licensing fix while the litigation is ongoing is itself evidence that the executive does not regard the fair dealing question as closed.
5. The global picture shows the same uncertainty, not a uniquely Indian gap. In the United States, The New York Times’ suit against OpenAI and Microsoft, filed in December 2023, remains in contested discovery in 2026, with no court having yet established that AI training on copyrighted journalism is fair use as a general matter. India’s unsettled position sits within a genuinely unresolved global debate, not behind it.
6. The commercial interests on both sides deserve scrutiny. Publishers seeking stricter licensing terms, including DNPA members, have an obvious commercial stake in that outcome, just as OpenAI has a stake in a permissive reading. Neither interest makes the underlying legal question resolved; both are reasons to wait for the trial and for legislative clarity rather than treat either side’s preferred reading as the default.
Data and Institutions Vault
Prelims-grade facts:
- ANI Media vs OpenAI: suit filed in Delhi High Court, November 2024
- Interim order denying ANI’s injunction application: 24 July 2026, by Justice Amit Bansal
- Statutory basis invoked, prima facie: Section 52(1)(a), Copyright Act, 1957 (fair dealing)
- Infringement provision: Section 51, tied to exclusive rights under Section 14
- EU’s text and data mining exception: Article 4, Digital Single Market Directive (Directive 2019/790), opt-out based
- India’s parallel licensing proposal: DPIIT, December 2025, mandatory royalty framework for AI training data
- US comparator case: The New York Times vs OpenAI and Microsoft, filed December 2023, in discovery as of 2026
- Publishers’ body seeking to join proceedings: Digital News Publishers Association (DNPA)
Watch the trap: do not write that the Delhi High Court has “ruled that AI training on news content is legal in India.” It denied an interim injunction on a prima facie, provisional reading of fair dealing; the suit is still pending, and no final finding on infringement has been made.
The Debate
Argument FOR requiring licences before AI training. Fair dealing under Section 52 is a narrow, enumerated exception never intended to cover the wholesale, automated ingestion of an entire news archive by a commercial model. Without an explicit text and data mining exception, extending fair dealing this far by judicial interpretation alone, at the interim stage, risks eroding the commercial value of journalism before Parliament or a full trial has weighed in.
Argument AGAINST treating licensing as the default answer. Fair dealing for research and reporting has historically been read broadly, AI-driven discovery can plausibly benefit publishers through visibility, and a low-friction training regime may be necessary for a genuinely transformative technology to develop in India rather than only in jurisdictions with more permissive rules. Requiring bespoke licences for every use before any final ruling exists could pre-empt a legal question that remains genuinely open.
Balanced verdict. The strongest position is procedural, not substantive: neither side should treat the current state of Indian law as settled. The interim order is real and reasoned, but it decided an injunction application, not the suit. Until the trial concludes or Parliament legislates a clear text and data mining exception or licensing framework, AI developers operate at genuine legal risk if they treat the prima facie fair dealing finding as a general licence, and publishers overstate their position if they claim the law already requires licences in every case.
How to Think About This
The transferable pattern: an interim court order resolves the narrow question put before it, for the specific relief sought; it does not resolve the broader legal or policy question merely because commentary treats it as though it had.
Interim orders exist precisely because full trials take time and some questions cannot wait. Courts routinely make prima facie findings, assessments made on the evidence available at that stage, sufficient to decide whether interim relief is warranted, without finally deciding the underlying dispute. The common analytical error is to read the outcome of the interim application (injunction granted or denied) as though it were the outcome of the full case, when the two can diverge substantially once complete evidence and argument are before the court.
This same structure recurs whenever a stay is vacated in a constitutional challenge without the underlying law being upheld on merits, whenever bail is granted without the accused being acquitted, and whenever a regulatory body’s show-cause notice is stayed without the underlying regulatory question being settled.
Diagram-in-Words
WHAT THE INTERIM ORDER ACTUALLY DECIDED
ANI files suit (Nov 2024)
│
▼
ANI seeks INTERIM INJUNCTION to stop OpenAI's training/use
│
▼
Delhi HC (24 July 2026): INJUNCTION DENIED
│
├── prima facie: training may fall within fair dealing (Sec 52(1)(a))
├── prima facie: RAG output not substantially similar to ANI's reporting
└── jurisdiction upheld (OpenAI's objection rejected)
│
▼
THIS IS NOT: "AI training on Indian news content is lawful" (final finding)
THIS IS: "ANI has not shown enough, at this stage, for an injunction"
│
▼
SUIT CONTINUES toward full trial ── final infringement question still open
│
▼
PARALLEL TRACK: DPIIT licensing/royalty proposal (Dec 2025) ── still no
statutory text-and-data-mining exception in India (unlike EU Article 4)
India's Copyright Act, 1957 has no explicit exception for text and data mining, unlike the European Union's Digital Single Market Directive. Critically examine the implications of this legislative gap for AI training on copyrighted news content, with reference to the Delhi High Court's interim order in ANI Media vs OpenAI. (250 words)
Training a large language model requires ingesting very large volumes of text, and Indian news archives are a valuable, high-quality source of that text. The legal question is whether such ingestion is copyright infringement under Section 51 of the Copyright Act, 1957, or falls within a fair dealing exception under Section 52.
The Delhi High Court’s interim order of 24 July 2026 in ANI Media vs OpenAI, which denied ANI an injunction and found prima facie that LLM training could fall within fair dealing, has brought this question to the centre of Indian AI policy.
The order itself is narrower than headlines suggest. Justice Amit Bansal’s findings, that training falls within fair dealing under Section 52(1)(a), that ChatGPT’s generated outputs were not substantially similar to ANI’s reporting, and that the Delhi High Court has jurisdiction, were made for the limited purpose of deciding whether to grant interim relief, not as a final ruling on the suit, which continues toward trial.
Section 52’s enumerated exceptions, private or personal use including research, criticism or review, and current-events reporting, were not written with mass, automated ingestion of an entire news archive by a commercial model in mind, and India, unlike the European Union under Article 4 of the Digital Single Market Directive, has no dedicated, opt-out based exception for text and data mining. That the Department for Promotion of Industry and Internal Trade proposed a mandatory licensing and royalty framework for AI training data in December 2025, while the litigation was ongoing, shows the government itself does not treat the fair dealing question as closed.
The comparable US litigation, The New York Times’ suit against OpenAI and Microsoft, remains in contested discovery in 2026 with no settled fair use finding, underlining that this is an open question globally, not one where India is uniquely undecided.
The honest reading holds two things at once: the Delhi High Court’s order is a genuine, reasoned setback for ANI’s injunction bid, and it is simultaneously not proof that AI training on Indian news content is lawful in general. Durable clarity requires a legislative or regulatory answer, an explicit text and data mining exception or a working licensing and royalty mechanism, rather than reliance on a single interim order that the Court itself described as provisional.
Until that exists, both publishers and AI developers are better served treating the current legal position as unsettled rather than resolved.
Interview Angle
"An interim order that denies an injunction on a prima facie reading of fair dealing is not the same as a final ruling that the underlying use is lawful. Why does that distinction matter so much for how a licensing market for AI training data should be built, and what happens if industry treats the interim reading as settled law before the trial concludes?"
Takeaway Box
Lift line for an answer:
The Delhi High Court refused to stop OpenAI. It did not say OpenAI was right.
Prelims hooks: ANI vs OpenAI suit filed November 2024; interim order 24 July 2026 (Justice Amit Bansal); statutory basis Section 52(1)(a), Copyright Act 1957; EU TDM exception Article 4, DSM Directive; DPIIT licensing proposal December 2025; US comparator NYT vs OpenAI and Microsoft (filed December 2023).
Ethics and interview angle: when a court’s provisional, appealable finding gets reported and repeated as a settled legal rule, who bears the cost of that misreading, the publisher whose content was used, the AI company that relied on it in good faith, or the public that receives confident but premature legal claims as fact?
PYQ linkage: UPSC has repeatedly tested intellectual property rights under GS3 and government policy design under GS2; this editorial updates both with a live, named 2026 case at the intersection of copyright law and artificial intelligence, a combination increasingly likely to appear in future papers.
Probable question: “An interim order denying injunctive relief on a prima facie reading of a statutory exception is often mistaken for a final ruling on the underlying legal question.” Discuss with reference to copyright and AI training in India.
Sources: Hindustan Times, PIB
The Delhi High Court’s interim order of 24 July 2026 in ANI Media vs OpenAI, which denied ANI an injunction and found, on a prima facie basis, that training a large language model on its content could fall within the fair dealing exception under Section 52(1)(a) of the Copyright Act, 1957, is not a final declaration that AI training on Indian news content is lawful; the editorial argues that AI developers must not read a provisional, appealable finding made only for the purpose of refusing interim relief as settled licence to scrape news content without permission, and that the absence of any explicit text and data mining exception in Indian copyright law, unlike the European Union’s opt-out based Article 4 exception, leaves the underlying legal question genuinely open until the suit is finally decided or Parliament legislates.
- The procedural facts matter precisely because they are limited. ANI filed suit against OpenAI in the Delhi High Court in November 2024 alleging unauthorised use of its news content to train ChatGPT. On 24 July 2026, Justice Amit Bansal denied ANI’s application for an interim injunction, holding prima facie that large language model training could fall within fair dealing, that ChatGPT’s Retrieval Augmented Generation outputs were not substantially similar to ANI’s original reporting, and that the Delhi High Court had territorial jurisdiction over the dispute.
The Court was explicit that these were prima facie observations for the limited purpose of the injunction application, not a final finding on the merits, and the underlying suit continues toward full trial. - The statutory architecture is narrow by design. Section 51 of the Copyright Act, 1957 makes unauthorised exercise of an author’s exclusive rights under Section 14, including reproduction, an infringement.
Section 52 carves out fair dealing exceptions for private or personal use including research, criticism or review, and the reporting of current events, none of which was drafted with mass, automated ingestion of an entire news archive by a commercial AI model in view. Reading such large-scale training into a decades-old, narrowly enumerated exception is precisely the kind of extrapolation the editorial cautions against treating as settled. - India’s own policy apparatus appears to agree the copyright framework is inadequate as it stands. The Department for Promotion of Industry and Internal Trade circulated a proposal in December 2025 for a mandatory licensing and royalty framework for AI training on copyrighted content, including a proposed collecting body to distribute payments to rights holders.
A government department drafting a licensing solution in parallel with active litigation is itself evidence that the fair dealing question is not considered closed even inside the executive. - The comparative picture reinforces the same caution. The European Union’s Digital Single Market Directive gives text and data mining a dedicated, opt-out based exception under Article 4, covering commercial use unless a rights holder reserves its rights through machine-readable means, a purpose-built legislative answer India does not have.
In the United States, The New York Times’ copyright suit against OpenAI and Microsoft, filed in December 2023, remains in contested discovery in 2026 with no fair use finding yet established by any court, showing the same underlying question is unresolved well beyond India.
A more permissive reading has real force. Fair dealing for news reporting and research has historically been read broadly in Indian courts, and an efficient, low-friction AI training regime arguably benefits both publishers who gain visibility through AI-driven discovery and a genuinely transformative technology that could be stifled by requiring bespoke licences before every use.
It is also fair to note that the publishers pressing for stricter licensing terms, including members of the Digital News Publishers Association who have sought to join the proceedings, have a direct commercial stake in that outcome, which should temper any assumption that the licensing-first position is simply the neutral or self-evidently correct one.
India needs statutory clarity rather than reliance on a single interim, appealable order to govern an entire industry. That could take the form of an explicit, EU-style text and data mining exception with attribution and compensation built in, or a clearer licensing and royalty mechanism along the lines DPIIT has already proposed.
Until either exists, AI developers should treat the Delhi High Court’s prima facie fair dealing finding as provisional rather than a general licence, and pursue direct licensing arrangements with Indian publishers, as several global AI companies have already done with news organisations elsewhere, rather than wait for a final verdict that could still go the other way.
Cite this page: No News Content for AI Training Without Licences: Reading the ANI vs OpenAI Order Correctly — Ujiyari.com | Free UPSC & State PCS Editorial Analysis
