OpenAI and Microsoft press their fair-use case against The New York Times

OpenAI, Microsoft and The New York Times have presented their summary-judgment arguments in the newspaper’s copyright lawsuit. The central question is now sharply drawn: can AI companies train on published journalism as a transformative use, or does that copying help create products that compete with the original work? A ruling could reshape both model development and publishers’ ability to control and charge for their archives. It will not settle every AI copyright dispute, but it may put a very large road sign beside the industry’s favourite legal shortcut.

OpenAI Watch analysis

What happened

The parties filed their arguments with US District Judge Sidney Stein ahead of a possible trial. According to the Axios report, the judge is expected to decide in the coming months whether the case, or key parts of it, should proceed to trial next year.

The Times says OpenAI and Microsoft copied its journalism at scale to build commercial substitutes. OpenAI argues that established copyright principles permit transformative training on publicly available works, that reported facts are not protected by copyright and that studies indicate ChatGPT answers do not substitute commercially for Times articles. Microsoft advances a similar position for Copilot. These are the parties’ arguments as reported by Axios; WittyWires has not independently reviewed the filings.

The central claims

• The Times alleges market substitution
  It says AI products built with copied journalism can compete with the original publisher, making this more than an abstract training dispute.
• OpenAI argues training is transformative
  Its position is that turning millions of works into a general-purpose model delivers a different function permitted by fair-use principles.
• Outputs matter as much as inputs
  The fight turns partly on whether ChatGPT and Copilot answers replace journalism in the market, not simply whether articles entered training datasets.
• Earlier AI rulings are contested precedent
  OpenAI cites California decisions favouring Meta and Anthropic, while the Times says their facts differ and they do not bind a New York court.
• The US government backs OpenAI’s position
  The Justice Department argues that large-scale model training provides transformative public benefit outweighing competitive harm to publishers.

Why it matters

The practical stakes extend well beyond three famous names. A Times victory could strengthen publishers’ leverage to license archives and challenge unlicensed training. A victory for OpenAI and Microsoft could give model developers firmer ground for treating large-scale ingestion as fair use, particularly where outputs are not shown to replace the source.

The distinction between training and substitution is the hinge. If courts demand evidence of competing outputs, publishers may need to show market harm model by model. If copying alone carries more weight, the economics of assembling training corpora could change rather abruptly.

Our read

This is the useful legal phase, not merely another exchange of press statements. Readers should watch the judge’s treatment of market substitution and precedent: those points will reveal whether recent AI fair-use victories travel cleanly into a dispute involving journalism, or arrive with their luggage missing.

Neither side has won. Summary-judgment arguments are advocacy, and the existence of prior favourable rulings does not predetermine this case.

What to watch

• Whether Judge Stein resolves major claims or sends the central questions to trial.
• How the court measures competition between AI answers and original journalism.
• Whether the California Meta and Anthropic decisions influence the New York analysis.
• Whether any ruling changes licensing negotiations between publishers and AI companies.

Discussion spark: Should fair use for AI training depend mainly on how the model was built, or on whether its outputs compete with the original publisher?

OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.