OpenAI; The New York Times Company

OpenAI's February 2024 court filing turned The New York Times copyright case into a fight over evidence construction, not only training data. The Times had sued Microsoft and OpenAI in December 2023, alleging that generative AI systems used Times journalism without permission and could reproduce or substitute for parts of its work. OpenAI's partial motion to dismiss answered with a narrower but volatile claim: the Times' examples did not show ordinary product use, because they allegedly required repeated attempts, article excerpts, a bug and prompts OpenAI said violated its terms. The legal signal is the pressure this dispute puts on three control surfaces at once: publisher licensing leverage, AI training defenses and the evidentiary methods used to prove model-output copying.

The anchor is OpenAI's 26 February 2024 motion in The New York Times Company v. Microsoft Corporation et al, the Southern District of New York copyright case filed two months earlier. The Times' complaint alleged that Microsoft and multiple OpenAI entities used Times journalism without permission to train generative AI systems and that ChatGPT or related products could produce output that competed with, summarized or reproduced Times work. OpenAI's filing sought dismissal of several claims and challenged how the Times built its output examples.

OpenAI's central public move was to reframe the examples as adversarial evidence rather than normal user behavior. The motion alleged that the Times needed tens of thousands of attempts, article excerpts, a model bug and deceptive prompts to generate the outputs in the complaint. OpenAI also argued that ChatGPT is not a substitute for a Times subscription, that ordinary users cannot call up Times articles at will, and that the case should focus on viable copyright questions rather than what OpenAI viewed as contrived outputs.

The Times' side is not just a newspaper complaint about training data. Its complaint describes a business and rights conflict: expensive journalism used to build AI systems, outputs that may reproduce or closely summarize protected work, and a risk that users receive Times-derived information without visiting or paying the publisher. Axios later reported the Times' counsel response that OpenAI was mischaracterizing the publisher's use of OpenAI products to look for evidence of copying.

The operating importance is evidence discipline. If courts accept that output examples are too engineered, publishers may need more transparent test protocols and cleaner proof of ordinary-user substitution. If courts accept the examples as probative, AI providers face stronger pressure around memorization controls, prompt-attack resistance, content licensing and discovery into training data. Either way, the dispute raises the cost of treating model-output screenshots as simple proof.

The April 2025 order shows why the filing still matters beyond the headline. The court narrowed some theories but allowed important copyright-related claims to continue, including direct-infringement and contributory-infringement theories. That procedural context keeps the February 2024 allegation relevant as one part of a larger litigation path, not as a resolved finding that either side was right on the merits.