An Indian court just called AI training 'private use', and publishers should read the small print
The Delhi High Court refused news agency ANI an injunction against OpenAI, finding that storing copyrighted articles to train ChatGPT likely falls under a fair-dealing exception. It is an interim ruling, not a final verdict.
India’s first substantive court decision on AI training data has gone OpenAI’s way. In ANI Media v. OpenAI OpCo, pronounced on 24 July and widely reported this week, Justice Amit Bansal of the Delhi High Court refused the news agency ANI an interim injunction, holding that OpenAI’s storage of ANI’s articles for training the models behind ChatGPT prima facie falls under Section 52(1)(a) of India’s Copyright Act, the exception for “private or personal use, including research”. If that holds, it is not infringement.
Two details make the ruling more interesting than the headline. First, the court explicitly rejected ANI’s argument that a commercial company cannot claim that exception. Parliament, the judge reasoned, had limited certain other copyright exceptions to non-commercial use and had chosen not to do so here. Second, ANI actually won the jurisdictional fight. OpenAI had argued that Indian courts had no business hearing a case about training that happens on servers in the United States; Bansal disagreed, noting that accepting that logic would let any infringer escape Indian law simply by hosting abroad. The court also weighed public interest, warning that an injunction at this stage would harm the development of AI in India, including homegrown models, and affect millions of ChatGPT users there.
Here is the part that gets lost in the excitement. This was an application for an interim injunction, which is a request to stop something while the real case proceeds. The standard is whether the claimant has shown a strong enough case to justify freezing things now, not whether they will ultimately win. The main trial, with discovery and expert evidence, is still ahead. “Prima facie” is doing real work in that sentence: it means “on first look”, not “settled”. Even so, courts do read each other, and this is now the third significant ruling in about thirteen months to land on the side of AI developers on the training-data question, which is why publishers are treating it as a bad week rather than a technicality.
Worth knowing before drawing conclusions: this is Indian law, and the “private or personal use, including research” wording has no direct equivalent in EU or German copyright law, where the relevant provisions are the text-and-data-mining exceptions of the DSM Directive with their opt-out mechanism. A win in Delhi does not transfer to Munich.
What this means for you: If you are simply using AI tools, nothing changes; nobody is coming for your chat history. If you publish anything online, from a blog to a company knowledge base, the practical read is that courts in several jurisdictions are so far reluctant to treat training as infringement, so technical measures such as robots.txt rules and the EU’s TDM opt-out are doing more work than litigation right now. And if you build with AI, the legal ground under training data is genuinely still moving. These are interim rulings in unfinished cases, and treating any of them as permission would be premature.
Sources
Kimi K3's weights are finally public, and they weigh 1.4 terabytes
Moonshot AI has published the full weights and technical report for its 2.8-trillion-parameter Kimi K3, along with parts of the infrastructure needed to run it. The claim worth checking: 2.5 times more intelligence per unit of compute.
