The Justice Department supports OpenAI in the dispute over copyright with publishers.

      The Justice Department submitted a statement of interest in support of OpenAI’s fair use defense in the ongoing copyright cases with publishers, asserting that licensing fees could jeopardize national security. Unlike the U.S., EU law does not include a fair use doctrine, and the AI Act dictates copyright responsibilities for any provider offering a model in the Union market, regardless of where training took place.

      The U.S. government has informed a New York court that using publishers’ materials to train AI models constitutes fair use. This information was reported by CNET on Tuesday.

      The rationale presented centers on national security. The filing states that regulations making it more difficult to establish a strong American AI industry "threaten national security and confer a competitive edge to foreign adversaries who do not face such restrictions."

      The statement further argues that the costs of licensing would create an oligopoly of model training among the largest tech companies.

      In response, The New York Times has disagreed, with spokesperson Graham James stating that the administration is supporting a small group of trillion-dollar AI firms and asserting that AI companies should fairly compensate for the content essential to their products.

      These cases are consolidated in the Southern District of New York. The Times filed a lawsuit in December 2023, Ziff Davis (which owns CNET) followed with its suit in 2025, and 400 local newspapers have initiated their own legal actions.

      This reasoning is absent in Europe, where EU copyright law does not recognize a fair use doctrine and provides only a limited set of exceptions.

      Instead, Europe has a text and data mining exception with an opt-out provision, allowing rightsholders to reserve their rights under Article 4(3) of the 2019 copyright directive, which then nullifies the exception.

      The AI Act directly connects to this mechanism. Article 53 mandates that every provider of general-purpose models implement a policy that acknowledges and honors these reservations.

      This policy is associated with the model rather than the training process. Recital 106 specifies that any provider marketing a model in the Union must adhere to these rules, no matter where the copyright-related actions occurred.

      The recital emphasizes that no provider should receive a competitive advantage in the Union market by adhering to lower copyright standards than those established in Europe.

      This aligns with the Justice Department’s argument but in reverse. A Munich court ruled against OpenAI in November, determining that lyrics retained by GPT-4 constituted reproduction and that the text mining exception did not apply in this case.

      That ruling is currently under appeal and could be brought before the Court of Justice, based on actions that the Justice Department is now claiming are protected.

      Thus, a victory in Manhattan would not have implications in Europe, where OpenAI would face similar obligations in a market where publishers have already accused it of withholding evidence.

      Published September 2, 2026 - 10:10 pm UTC

      Back to top