AI copyright news, graded by source

Magistrate judge approves a protocol under which Midjourney must copy all image and video data used to train its models, v4 through v8, into cloud storage controlled by the studios' lawyers

On Sept. 29, 2026, Magistrate Judge A. Joel Richlin approved a joint stipulation that governs how Midjourney will produce its training data to Disney, Universal, Warner Bros. and the other studio plaintiffs. Under the order, Midjourney must identify every storage location holding the image and video datasets used to train its image models v4 through v8 and its video model, confirm in writing that those locations hold all of the data, and transfer a complete copy, including captions, metadata and retained embedding vectors, into Google Cloud buckets controlled by the studios' outside counsel and experts, isolated from Midjourney. It must also serve a manifest and a declaration under penalty of perjury from a named person, who can be deposed, stating that nothing was excluded. The sides split transfer and storage costs. The order implements a Rule 33(d) election Midjourney made on Nov. 21, 2025 and an Aug. 7, 2026 minute order, and says compliance satisfies Midjourney's duty to produce the datasets. Midjourney's July 2026 motion for judgment on the pleadings on the contributory infringement claim, citing the Supreme Court's Cox decision, remains pending before Judge Kronstadt.