A federal judge has denied music publishers’ motion for a preliminary injunction against AI giant Anthropic. Photo Credit: Igor Omilaev
A federal judge has denied music publishers’ push for a training-related preliminary injunction against AI giant Anthropic.
Judge Eumi Lee rejected the injunction request in a 13-page order, following considerable back-and-forth between the plaintiff publishers and the Claude developer. Among other things, the injunction sub-dispute saw Anthropic defend its existing training safeguards for protected works.
Ultimately, late 2024 delivered a training “guardrails” compromise of sorts between the parties. As we covered at the time, this partial injunction resolution settled questions concerning possibly infringing outputs from current and future Claude models.
But it didn’t resolve the second component of the music publishers’ sought injunction, involving Anthropic’s training of LLMs.
“Anthropic…[is] enjoined and restrained from using copies of lyrics (or portions of lyrics) to compositions owned or controlled by Publishers for future training of Anthropic’s AI models,” the plaintiffs wrote in their proposed injunction order.
Regarding the injunction’s scope – seemingly a big hang-up for the court – the publishers said they’d supplement the lawsuit’s 500 allegedly infringed works with a more comprehensive list. The latter would then update “on at least a quarterly basis,” per the text.
In the end, on the heels of related hearings and more, the court found the publishers’ “vague and unwieldy injunction” request to be “elusive and poorly defined.”
“Publishers’ counsel could not say how many songs would be subject to the injunction. … Publishers did not offer a concrete or definitive way for Anthropic – as the party subject to the injunction and the legal repercussions of a violation – to ascertain its parameters or comply with its terms,” Judge Lee indicated.
“The enormous and seemingly ever-expanding scope of Works included in the requested injunction raises significant concerns regarding enforceability and manageability,” the court drove home.
As for the possibility of issuing a more limited injunction (like for the above-noted 500 works identified in the suit), the judge emphasized that the publishers hadn’t “requested this narrower form of relief.”
“The Court therefore finds it inappropriate to issue a more limited injunction,” Judge Lee proceeded.
Perhaps more problematic – at least for music rightsholders when it comes to the broader infringement battle against gen AI developers – are the court’s findings pertaining to the alleged infringement’s resulting harm.
Of course, this position ties directly to the injunction motion, not the central copyright claims. Admittedly, though, its tone doesn’t appear to bode well for the publishers’ suit.
“But they [the publisher representatives’ declarations] do not demonstrate how using the Works to train Claude is affecting – let alone diminishing – the value of any of the Works,” the court wrote.
“Even if Publishers had shown that the use of the Works to train Claude will cause them harm within the emerging AI licensing market,” the judge found on the market-harm front, “they once again fail to show that such harm is irreparable.”
In a statement, the music publishers reiterated their continued confidence in the overarching suit.
Content shared from www.digitalmusicnews.com.