When Journalism Became ANI’s Weakness Against OpenAI

There is an unusual irony at the centre of ANI’s copyright battle against OpenAI. The news agency went to court arguing that its journalism had been appropriated to train and operate ChatGPT. Yet, when the Delhi High Court declined to grant it an interim injunction, some of the strongest reasons for refusing relief came from the nature of journalism itself.

News is built around facts. Facts cannot be owned. Events are reported by multiple organisations. Stories evolve as new information emerges. Journalists may have copyright over the language, structure, and creative choices through which they present information, but not over the information itself.

These familiar principles of journalism became critical legal considerations in Justice Amit Bansal’s ruling dismissing ANI’s application for immediate restrictions on OpenAI.

The court did not find that news reports are incapable of copyright protection. On the contrary, it accepted, at least at the interim stage, that ANI could own copyright in original literary works produced by its journalists and other professionals.

But copyright, the court stressed, protects the expression of facts, not the facts themselves. For ANI to secure an injunction, it was not enough to show that ChatGPT had generated responses about the same events, interviews or people covered in its reports. It had to demonstrate that the artificial intelligence system had reproduced a substantial part of ANI’s particular form of expression.

That distinction proved difficult for ANI to overcome.

The court examined the ChatGPT responses presented by the agency and found that, on a prima facie view, they did not materially reproduce ANI’s language. They conveyed information, summarised events or answered questions, but were not shown to copy the protected expression of the agency’s reports to the degree required for copyright infringement.

This was particularly important because journalism rarely exists in isolation. A political statement, a court order, a sporting achievement or an interview may be reported by dozens of organisations. The underlying facts may be identical even when every report is written differently.

Similarity in information, therefore, does not necessarily indicate copying. Two reports may refer to the same event, use the same names, dates and quotations, and arrive at similar summaries without one reproducing the other’s protected expression.

The judgment effectively recognised the cumulative character of news production. Journalism is rarely a single, sealed work. It is a continuing public conversation built from official releases, speeches, interviews, eyewitness accounts, data and earlier reports.

That made it harder for ANI to claim that a ChatGPT response containing similar information must have originated from, or infringed, an ANI report.

The agency had also alleged that OpenAI’s models memorised and regurgitated its work. But here, too, the evidence fell short at the interim stage.

The 14 reports cited in ANI’s original claim were published in August and September 2024. OpenAI argued that the relevant models had training cut-off dates of April 2022 for GPT-4 and April 2024 for GPT-4o. On that basis, the reports could not have been part of the training data for those models.

ANI did not produce sufficient alternative examples to establish that its reports had been memorised during training and reproduced in ChatGPT’s responses. The court therefore found no prima facie case of regurgitation.

The ruling leaves open the possibility that an artificial intelligence model could infringe copyright if it reproduces a news report verbatim or in a substantially similar form. Memorisation is not legally irrelevant. But it must be demonstrated through specific evidence, rather than inferred merely because an AI response contains the same facts as a published story.

Public Interest

The public function of journalism also shaped the court’s approach.

ChatGPT, the judgment observed, is used by people to obtain information, conduct research, translate material, summarise documents and learn about events. Restricting its operation at an interim stage would affect not only OpenAI but also users who rely on the service.

This introduced a wider public-interest question. Journalism claims social value partly because it distributes information. But that same value can weaken attempts to impose broad restrictions on other systems that also help users access and understand information.

The court was unwilling to grant a remedy that could disrupt access to information before ANI had clearly established infringement.

It also found no strong evidence that ChatGPT was replacing ANI’s business.

ANI is principally a news agency. It gathers information and supplies reports, photographs and video to subscribers. ChatGPT is a general-purpose technology offering a far wider range of functions. Even when it answers questions about the news, it generally provides summaries rather than acting as a full substitute for a news-agency feed.

The decision is not a final victory for OpenAI. The court’s findings are preliminary and confined to ANI’s request for interim relief. Questions concerning the legality of AI training, the limits of fair dealing and the circumstances in which an output becomes a substantial reproduction will continue to be examined during the trial.

But the immediate ruling exposes a difficult tension for the news industry.

Publishers argue that their reporting is valuable because it is original, expensive to produce and commercially licensable. At the same time, journalism derives its democratic legitimacy from the proposition that facts and information must circulate widely.

In ANI’s case, the second proposition constrained the first.

The agency sought protection for its journalism. The court responded by asking what, precisely, in that journalism could be exclusively owned. Its answer, for the moment, was narrow: the original expression may belong to ANI, but the facts, events and public information at its core do not.

Journalism was not declared unprotected. But its most essential feature—the communication of facts to the public—became one of the reasons the court refused to stop ChatGPT.