AI copyright news, graded by source

Midjourney's answer to Disney and Universal admits its models were trained on billions of publicly available images and argues that training is 'quintessentially transformative fair use'

Disney, Marvel, Lucasfilm, Twentieth Century Fox, Universal and DreamWorks sued Midjourney on June 11, 2025, alleging it trained its image generator on their films and lets users produce images of characters such as Darth Vader, Shrek and the Minions. In its Aug. 6, 2025 answer, Midjourney states in its preliminary statement that "Midjourney's AI models had to be trained on billions of publicly available images" paired with text descriptions, and in response to paragraph 115(c) admits that it uses publicly available images to train its models and that the specifics of the process have not been publicly disclosed. It admits its CEO said version 6 was its third model trained from scratch, and that many subscribers use email addresses associated with the plaintiffs' companies. Midjourney denies infringement, pleads fair use, and argues that training a model to extract statistical information from copyrighted works is transformative, citing the June 2025 Bartz and Kadrey rulings. It does not admit that any specific Disney or Universal work was in its training data. The case was later consolidated with Warner Bros.' suit (No. 2:25-cv-08376) before Judge John A. Kronstadt.