June was the month AI-generated media ended up in court.
Disney and Universal v. Midjourney
On June 11, Disney and Universal sued Midjourney for copyright infringement.¹ They say it trained on their films and lets users generate characters like Darth Vader and the Minions on demand. It is the first time major Hollywood studios have sued an AI image company.
Two Rulings on Training Data
On June 23, Judge William Alsup ruled in Bartz v. Anthropic that training Claude on books Anthropic bought and scanned was fair use. He also ruled that keeping millions of pirated books in a central library was not, and ordered a trial on that part.² It is the first US ruling on fair use for training a large language model.
Two days later, Judge Vince Chhabria ruled for Meta in Kadrey v. Meta, a similar case brought by authors.³ He was explicit that the ruling is narrow. The authors lost because they made the wrong arguments, and the decision does not mean Meta's use of their books was lawful.
Put together, the line is starting to take shape. Training on legally acquired works looks defensible. Building a library of pirated ones does not. And Chhabria's opinion points to the argument that could still win: AI output flooding the market for the original work.
Provenance Over Detection
Realistic AI video has been everywhere since Google released Veo 3 in May. Tools that try to detect AI-generated content after the fact have a poor track record. The more durable approach is provenance: attach signed metadata when content is created, and keep it intact as the content moves across platforms. The EU AI Act will require labeling for some AI-generated content, but those rules don't apply until August 2026.
For Product Teams
- Know where your data came from. "We bought it" and "we scraped it" now carry different legal risk.
- Plan for provenance and labeling if your product generates media.
- Filter outputs for well-known characters and brands. The Midjourney suit shows the risk of not doing it.
How is your team thinking about training data and provenance? I'd like to hear how others are handling it.