AI, News and Copyright: What Delhi’s ChatGPT Ruling May Mean Closer to Home

A Delhi High Court judgment on OpenAI’s use of news content to train ChatGPT gives an early look at how courts may approach AI and copyright and a useful lens for testing Malaysia’s own Copyright Act 1987.

The Backstory

ANI Media Pvt. Ltd. (“ANI”), a major Indian news wire agency, sued OpenAI Opco LLC (“OpenAI”), the company behind ChatGPT. ANI made two related complaints:

• Training claim: OpenAI copied and stored ANI’s news articles, without a licence, to train the large language models (LLMs) behind ChatGPT.
• Output claim: When users asked ChatGPT questions, its answers sometimes reproduced ANI’s original articles.

ANI had, before filing suit, offered OpenAI a licence to its content for USD 7.5 million. OpenAI did not take up the offer. ANI then asked the Court for an interim injunction—essentially, an order stopping OpenAI from using ANI’s content and requiring deletion of ANI’s data from OpenAI’s systems—while the main case was still being tried.

 

The Questions Before the Delhi Court

To decide whether to grant that interim order, the Court had to work through four questions:

1. Jurisdiction: Can an Indian Court hear a claim against a US company whose servers sit outside India?

2. The output claim: Does ChatGPT’s answers to users infringe ANI’s copyright?

3 & 4. The training claim and fair use: Does storing news articles to train an AI model, without a licence, infringe copyright, or is it protected as “fair dealing” under Section 52 of India’s Copyright Act, 1957?

 

What the Delhi Court Decided (For Now)

1. Jurisdiction: Yes, on a preliminary view

The Court held that it could hear the case. Training and the resulting output could not be treated as separate, unconnected acts. As the output reaches users in India, the fact that training servers sit abroad did not, at this stage, remove the case from the Court’s reach.

2. The output claim: No infringement made out, on the evidence so far

The Court found that ChatGPT’s answers were not a “substantial reproduction” of ANI’s articles. It noted that ANI’s own examples of alleged copying were drawn from articles published after OpenAI’s relevant training cut-off (so there was nothing to “memorise” from those specific articles), and that even deliberately worded, repeated prompts by ANI could not extract anything approaching a substantial copy.

3. & 4. Training and fair use: Protected, on a preliminary view

This is the heart of the judgment. The Delhi Court held that storing news articles to train the LLM fell within Section 52(1)(a)(i) of India’s Copyright Act, i.e. dealing with a work for “private or personal use, including research.” Two (2) points stand out:

• Commercial use was not fatal. The Delhi Court rejected the argument that “research” must be non-commercial. Reasoning: where India’s Copyright Act meant to exclude commercial use, it said so expressly elsewhere and Section 52(1)(a) did not.
• A broader “fairness” test was also satisfied. The Delhi Court found the use to be transformative, saw no evidence that it displaced ANI’s market or revenue, and weighed heavily the public benefit of AI tools in education, research and access to information.

ANI could not show a strong enough case to justify an injunction. Balance of convenience also favoured OpenAI: any injunction would disrupt a service used by millions of people in India, whereas ANI’s claim (unlike OpenAI’s potential loss) could be compensated later in damages if ANI succeeds at trial.

An important caveat

This was an application for an interim injunction, not the final trial. The judge expressly noted that his observations “would have no bearing on the final outcome of the suit.” The underlying case, on the merits, continues.

 

Reading This Against Malaysia’s Copyright Act 1987

Malaysia has its own fair dealing regime, and it is not identical to India’s. Two (2) features are worth noting for anyone tempted to assume the same result would follow here:

• The “non-profit” qualifier. Section 13(2)(a) of our Copyright Act 1987 protects fair dealing done for the purposes of “non-profit research, private study, criticism, review or the reporting of current events.” Unlike the Indian provision the Delhi Court applied, our “research” limb carries an express non-profit qualifier on its face. A commercial AI developer relying on a “research” argument in Malaysia would have to grapple with that wording. This is a hurdle the Delhi Court simply did not have to address, since India’s equivalent provision does not carry it.
• We already run a fairness-factor test. Section 13(2A), inserted in 2012, requires Malaysian Courts to weigh factors such as the purpose and character of the dealing, the nature of the work, the amount used, and the effect on the potential market. This is a hybrid “fair dealing/fair use” approach not unlike the “fairness test” the Delhi Court applied.

News articles would ordinarily qualify as literary works under Section 3 and reproducing them (including by storage in electronic form, for training purposes) would engage the reproduction right in Section 13(1)(a). This means that the fair dealing exceptions in Section 13(2)(a) and (2A) are exactly where an equivalent argument in Malaysia would need to be fought and won.

None of this tells us how a Malaysian Court would actually rule. It only tells us where the fight would happen, and that the outcome is genuinely open: a question of statutory interpretation, evidence and policy that only our Courts (or Parliament, through legislative reform) can ultimately settle.

 

What This Could Mean for Malaysian Businesses

The implications cut both ways, depending on where a business sits:

• Content owners and publishers may want to think ahead about licensing strategies, technical measures (such as “crawler restrictions”, i.e. rules and methods which allow, limit or block automated bots, scrapers and AI training systems from accessing digital content), and clear terms of use. Furthermore, content owners and publishers ought to keep records of any commercial harm AI tools may cause, since “effect on the market” is squarely part of the fairness analysis under our Act.
• Businesses building or deploying AI tools should be cautious about assuming a “research” or “fair dealing” defence will travel neatly from India to Malaysia, given the non-profit qualifier. Licensing arrangements, rather than relying on an untested statutory exception, may remain the safer commercial path for now.
• Any business using generative AI output commercially should bear in mind that even if training itself is protected, an output that closely mirrors someone else’s original work could still raise separate infringement risks similar to what the Delhi Court’s own output-claim analysis shows.

How Malaysian Courts would apply Sections 13(1), 13(2)(a) and 13(2A) to AI training and AI-generated output remains untested, and the answer will depend heavily on the specific facts of any future case.

 

Where This Leaves Us

Nothing in the Delhi ruling changes Malaysian law, and it may not survive final determination in India. But rights holders and AI developers operating here should not treat the silence of our statute as comfort. Until the position is tested, licensing terms, not litigation outcomes, remain the more reliable form of protection.

This article was authored by Suaran Singh Sidhu (Partner) and Ashwinathan Selvanathan (Associate).

---

Key contacts

Please contact our team if you have any questions.

Suaran Singh Sidhu| Co-Head, Intellectual Property