Indian judge lets AI training stand, and music rights holders are watching

An Indian judge has ruled that OpenAI was within the law when it used content owned by the news agency ANI to train ChatGPT without permission, holding that the training is covered by a private research exception in Indian copyright law. Judge Amit Bansal was deciding ANI's bid for an interim injunction, and refused it.

The reasoning matters more than the immediate result. Indian copyright law has no specific text and data mining exception of the sort AI companies usually reach for, so OpenAI's lawyers pointed instead at the exception covering private or personal use, including research. To many Indian rights holders that looked like a stretch.

ANI and its supporters argued that a private research exception should cover non-commercial uses and private individuals, not one of the most valuable companies in the world, and questioned whether model training counts as research at all. Bansal was unmoved. The equivalent UK exception is explicitly limited to non-commercial research; the Indian one is not. And private, he concluded, describes how the research is conducted rather than who conducts it.

On the facts, he found that ANI's content was stored in a closed space with no public access and was not available to any human for access or download, which in his view made the use purely private. Training that involves machine learning of stored literary works by screening, organising and analysing them, then converting them into machine-readable inputs, sounded to him like a form of research.

The decision concerns journalism, and a ruling about the written word does not automatically set a precedent for recorded music. The Indian record industry trade group IMI still made a submission to the court, which tells you how closely the music business is reading these cases.

There is a jurisdictional twist that could matter more than the exception itself. OpenAI actually trained ChatGPT in the United States and would argue its training is subject to US law, where fair use for AI training is the subject of dozens of unresolved lawsuits. If US courts eventually find training is not fair use, a country whose own law explicitly permits it becomes an attractive place to base training operations.

Rights holders counter that where a model is commercially exploited matters as much as where it was trained, though that position is disputed and may need new legislation to become enforceable. It is also harder to police against open-weight models that users download and run themselves. Elsewhere in the same fight, Sony Music has filed a fresh lawsuit against the AI startup Udio, and Luminate has found that most people want AI use disclosed and proposed a framework for tracking AI-generated music.