Indian judge rules ChatGPT training was legal under a private research copyright exception

An Indian judge has ruled that OpenAI did nothing unlawful when it trained ChatGPT on content owned by news agency ANI without permission. Judge Amit Bansal found the use fell inside a “private research” exception in Indian copyright law, and refused ANI the interim injunction it had asked for.

The case covers journalism, not music. The reasoning is what matters to the music business, because a court has now read an existing copyright exception wide enough to cover AI training without a single licence changing hands.

What Judge Amit Bansal actually ruled

Bansal split the question in two, and OpenAI won both halves.

On storage, he found the copied content never became public. It sat “in a closed space without access to the public,” he wrote, and was “not publicly available to any human entity either for access or for download.”

“

In my opinion, the use amounts to being purely private.

On training, he found the process looks like research. It “involves machine learning of the stored literary works by screening and organising them,” Bansal wrote, then analysing the data and “making extractions from the literary works and converting them into machine-readable training inputs.”

Private plus research equals the exception. ANI’s injunction, which would have blocked OpenAI from using its content in future training, was denied.

Why India’s missing TDM exception forced a different argument

Most AI copyright fights outside the US start with text and data mining. A TDM exception lets a company copy protected works to extract patterns from them, and the EU wrote one into law.

India has no TDM exception. With that door shut, the argument had to run through a provision written decades before anyone trained a model, and the judge stretched “private research” to cover it.

This is the pattern worth watching. In the US, AI companies argue fair use, a flexible standard. Everywhere else they need a specific carve-out, and when none fits, they ask a court to widen an old one. The UK government has refused to create a new mining exception under sustained lobbying. India’s courts did something close to it anyway.

Why Indian labels filed in a news agency’s case

IMI, the trade body for India’s recorded music industry, made a submission to the court in a dispute about news copy. That tells you how the industry reads the stakes.

A ruling about literary works sets no binding precedent for sound recordings or compositions. The reasoning travels, though. A judge who accepts that training is “research” and that a private corpus is “private” has built a frame any AI company can point at when the works are songs instead of wire copy.

IMI’s members include Sony Music India, Universal Music India, and T-Series. Their preferred outcome is a licensing route, the direction GEMA has taken in Germany. A court reading exceptions generously moves the ground the other way.

What the ANI ruling could mean for where AI models get trained

OpenAI trained ChatGPT in the United States, and would argue US copyright law governs the training regardless of what an Indian court says.

That defence cuts both ways. As CMU noted, if US courts eventually decide AI training is not fair use, a ruling like this one gives AI companies a reason to move training operations to India instead.

Rightsholders have a counter-argument ready: where a model gets commercially exploited should matter as much as where it was trained. A model trained legally in one country, on that logic, cannot be sold into a country where permission would have been required. That position is disputed in most places and is not clearly written into law anywhere, which is the gap the industry now has to close.