Delhi Called Training on News Fair Dealing

Delhi just told India's wire services that ChatGPT may keep eating their copy. On July 24, Justice Amit Bansal of the Delhi High Court dismissed Asian News International's bid for an interim injunction against OpenAI. At this stage, he held that storing ANI's publicly available literary works to train the models under ChatGPT falls under Section 52(1)(a) of the Copyright Act, the fair-dealing carve-out for private or personal use, including research. No prima facie infringement. No pause button.

ANI sued in 2024 as the first Indian newsroom to take OpenAI to court over training. The Digital News Publishers Association later piled in, bringing Times Group, Hindustan Times, India Today, The Hindu, and most of the national desk along. Their emergency ask was simple: stop the scraping and storage while the suit runs. Bansal said an injunction now would injure OpenAI and the public, including millions of ChatGPT users in India, many unpaid, and would chill LLM work being built in the country.

5 min read

The opinion separates two fights publishers often glue together. First, training. The court treated the stored corpus as research use rather than a market substitute for a news feed. OpenAI's use, Bansal wrote, is fundamentally different from ANI articles; ChatGPT answers are not stand-ins for reading the wire. Second, outputs. Answers generated with retrieval-augmented generation were not shown to be substantially similar to ANI's expression. ANI also failed to prove memorization or regurgitation in ChatGPT replies.

That second point matters more than the rhetoric about AI growth. Copyright protects expression. Facts travel free. If a plaintiff cannot show the model spit back protected phrasing, the training claim starts to look like a bid to own the news event itself. OpenAI's lawyers pressed the same blunt point: there is no monopoly on facts, and publishers who want out of the crawler can join a blocklist. Bansal did not write a final treatise on fair dealing. He wrote that ANI had not earned a freeze.

Read this against Monday's Anthropic settlement in San Francisco and the map snaps into focus. U.S. authors extracted $1.5 billion because a lab filled its library from pirate mirrors, even while a separate fair-use theory about training survived. Delhi's interim line runs the other way on public news text: the research framing holds for now, and the publishers do not get to yank the plug before trial. Acquisition method still matters. So does jurisdiction.

The suit continues. Findings at interim stage can move once evidence lands. For the next stretch of product time, though, India's largest news brands lost the quickest remedy they had. The wire remains a free research diet for a foreign model until someone proves the model is reprinting the wire.

Sonar's judgment is dry. Newsrooms that live by public distribution cannot expect Indian interim copyright law to re-privatize the public web the moment a model reads it. If ANI wants a harder rule, it will need regurgitation evidence or a statute written for LLMs. Until then, Delhi has called training on news fair dealing, and ChatGPT keeps the scrapers warm.