News publishers ask a judge to reject OpenAI's fair use defence, quoting Microsoft's staff

Five news organisations have asked a federal judge to rule that OpenAI and Microsoft broke copyright law in the way they built and run their AI products. The argument sits in a 92-page summary judgment brief filed on September 17 in the US District Court for the Southern District of New York, docket 1:25-md-03143, and every word of it is one side's case. OpenAI and Microsoft dispute the claims, are defending on fair use, and the court has ruled on none of it.

Copilot sits at the centre of the brief: the grounding and output counts are pressed against Microsoft alone, and the click-through data compares Copilot with Bing search. Image: Microsoft

The brief opens on a Microsoft executive's own words. It quotes Brent Hecht, the company's director of applied science, writing that "millions of people around the world will soon consider large models 'hoovering up' all their work to be an astonishing theft of unprecedented proportions", and, separately, that a win for the companies would arguably "make a complete mockery of the idea of 'fair use'". Both lines are the plaintiffs' selections from internal documents obtained in discovery, quoted to support their own motion.

Five publishers are asking for a ruling on five kinds of copying

The plaintiffs are The New York Times, the Daily News group of eight papers including the Chicago Tribune and the Denver Post, Ziff Davis, whose brands include IGN, CNET, PCMag, ZDNET and Mashable, the non-profit newsroom behind Mother Jones and Reveal, and The Intercept. Their five cases were consolidated into one multidistrict proceeding in New York and are now argued together.

The motion asks the court to find infringement at five points: acquiring the articles, training models on them, using Microsoft's Bing index to ground live answers, reproducing text in Copilot outputs, and what the filing calls "horse trading", the two companies passing copies to each other under their partnership. The grounding and output counts are pressed against Microsoft alone, which a footnote explains by pointing to an unresolved sanctions motion against OpenAI. The brief also asks for a ruling that OpenAI intentionally stripped copyright management information from the articles, and that statutory damages should be counted per article rather than in a lump.

The filing leans on the companies' own people

OpenAI's head of ChatGPT, Nick Turley, wrote that publishers face an "existential threat" from these products and that they "are largely substitutive, period", the brief says. It also says Satya Nadella agreed under oath that talking to a chatbot has "substituted" for going to the underlying source.

One Microsoft document quoted in the filing goes further, calling the situation a "doom loop" in which "an end-product threatens the economic foundations of its essential suppliers". These are quotations chosen by the plaintiffs to make their case, presented without the surrounding documents, and none of them has been weighed by the court.

The click-through numbers are the heart of the harm argument

Microsoft's own data, as the brief presents it, shows click-through rates falling 87 to 93% for The Times's sites, 83 to 91% for the Daily News sites and 51 to 94% for Ziff Davis sites when Copilot's answer engine is compared with traditional Bing search. The filing adds a third-party study finding that 87.78% of ChatGPT users visit no external website during a search, against 26.91% of Google users. None of it has been tested in court, and the defendants have not yet filed their response.

A cost figure runs alongside it. Using OpenAI's published API pricing, the brief calculates that generating one million news-style articles of about 500 words would cost roughly $6,800 and require no reporter or editor, which it offers as evidence that the products make low-quality news farms cheap.

Two specific accusations about OpenAI's handling

The brief says OpenAI obtained The New York Times Annotated Corpus, a dataset of more than 1.8 million articles published between 1987 and 2007, under a licence limited to non-commercial research, and trained on it anyway. It also says that immediately after the lawsuits were filed OpenAI built a "Bloom filter" to suppress the plaintiffs' content in outputs, while leaving content from publishers that had not sued unfiltered. Both are allegations in the plaintiffs' own document rather than findings, and OpenAI's answer to them is not in it.

Custom GPTs appear as evidence too, with the filing reproducing listings named "Bypass Paywall", "Remove Paywall" and "NYTimesGPT". That platform is on its way out: OpenAI is retiring custom GPTs and moving their functions into plugins.

None of this is a finding

A summary judgment brief is an argument for a judge, not a decision by one. The document is partly redacted, every quote in it is the plaintiffs' selection, and the fair use defence it attacks has not been ruled on in this case. OpenAI and Microsoft will answer with their own filings, and the judge may refuse the motion in full.

What is unusual is the witness list. The publishers' strongest lines here come from a Microsoft director, a Microsoft chief executive under oath and OpenAI's head of ChatGPT, which is why the filing is worth reading even before a judge says a word about it.