In October 2023, several large music publishers, Concord, Capital CMG, Universal and ABKCO, ( the “Plaintiffs”) brought copyright infringement proceedings against Anthropic PBC (“Defendant”) in the US, alleging the Defendant, an AI developer, had infringed on their rights by using their song lyrics catalogues to train its flagship AI product, “Claude”.
The Plaintiffs also sought a preliminary injunction ( a temporary injunction effective while the case proceeds through the Courts), seeking an order for the Defendant to “maintain effective guardrails” on its AI-generated output[1]; and restrain the Defendant from using the Plaintiff’s lyrics in AI training.
On 25 March 2025, US District Judge Eumi K. Lee denied the Plaintiffs’ application for an interim injunction restraining the Defendant’s use of the Plaintiff’s lyrics in AI training[2].
This decision has been hailed as an “early victory” for AI developers, but is it really a victory?
First, as Judge Lee pointed out in her ruling, under U.S. law, “a preliminary injunction is an extraordinary remedy never awarded as of right”.
Under U.S. law, a plaintiff seeking an interlocutory injunction “must establish that he is likely to succeed on the merits”. This is a threshold significantly higher than the American Cyanamid threshold under UK/Hong Kong law, where the Plaintiff only needs to show “there is a serious issue to be tried or there is an arguable case”.
It therefore appears successfully obtaining a preliminary injunction in the U.S. is always an uphill battle.
The Plaintiff’s problems were exacerbated by the difficulty in defining the scope of the proposed injunction, with the Judge outright stating “the details [for the proposed injunction] remain elusive and poorly defined” and the proposed injunction “vague and unwieldy”.
As the state of music catalogues is always in flux, with compositions and lyrics constantly moving in or out of a publisher’s catalogue, it is perhaps not surprising setting out a conclusive list is impossible.
It appears the lack of clarity on scope, which in turn led to concerns regarding enforceability and manageability, was a significant factor for the Judge in denying relief.
The main thrust of the Judge’s decision is the Plaintiffs could not establish “irreparable harm”, i.e. harm that cannot be remedied through financial compensation.
This could spark some interesting discussion.
The Plaintiffs argued the Defendant’s use of their copyrighted lyrics would “erode the value of the Works by undermining the licensing market, damaging [the Plaintiffs’] position to negotiate further training licenses with AI developers.”
It also “reduces demand for licenses for the lyrics, and undermines Publishers’ ability to negotiate new licenses and to renew existing licenses”.
The Judge apparently seized on these arguments and reasoned such “harm” could be dealt with by financial compensation, so there were no grounds on which to grant preliminary injunctions.
In denying the Plaintiffs’ application for preliminary injunction, the Judge devoted several paragraphs in discussing the “emergent market for AI training,” where copyright owners grant licences to AI companies to use their work in exchange for licence fees).
It should be noted, however, the “effect of the use upon the potential market” is precisely one of the four factors the U.S. Courts are to consider in deciding if a defendant’s use of a plaintiff’s copyrighted work can be protected by the defence of “fair use”.
As I have discussed in several earlier articles, “fair use” is the main defence all AI companies are running against copyright infringement claims in the U.S..
Judge Lee’s comments seem to highlight that a “market for AI training” does exist.
As such, AI companies’ use of copyright work as training materials could be undertaken by way of a licensing framework and their failure to obtain such licences for training will inevitably have an impact on the “market for AI training”.
While Judge Lee stressed she was not adjudicating on the issue of fair use[3], her analysis would seem not to bode well for the future of the fair use defence. Will this victory spell trouble for AI companies down the road?
Judge Lee’s comment that infringement could be adequately compensated by financial means also throws up some interesting questions.
If AI companies’ fair use defence does fail, what remedies would be available to copyright owners whose works have already been used to train AI?
Should AI companies be forced to take down their existing modules and “retrain” them with public domain and or licensed work?
Or would copyright owners be “forced” to accept some sort of “compulsory” licensing regime similar to RAND (Reasonable and non-discriminatory) or FRAND (fair, reasonable, and non-discriminatory) licensing regimes under patent laws?
Can copyright owners, particularly individual authors, refuse to license as a matter of principle?
Clearly, victory or not, the battle is a long way from over.
Disclaimer: This article is intended to provide general information only and does not constitute legal advice. Specific legal advice should be sought in relation to any specific situation. Haldanes, as a Hong Kong legal adviser, is qualified to advise on Hong Kong law only and does not offer opinions on the laws of other jurisdictions. Haldanes shall not be held responsible or liable for any losses or damages arising from the use of this article, including any errors or omissions, or any reliance placed on its contents. The reader assumes all risks and responsibilities associated with the use of this article.
For further information, please contact:
Anthony Leung, Partner, Haldanes
anthony.leung@haldanes.com
[1] This relief was not pursued after the Parties reached a Court-sanctioned agreement in December 2024 for the Defendant to maintain its already-implemented guardrails on all present and future Claude models, i.e. that the Defendant implements coding in the Claude chatbot to prevent it from providing lyrics owned by the Plaintiffs or creating new song lyrics based on copyright material.
[2] Decision available at: https://storage.courtlistener.com/recap/gov.uscourts.cand.431519/gov.uscourts.cand.431519.321.0.pdf
[3] The Judge wrote, “By seeking a preliminary injunction, Publishers are essentially asking the Court to define the contours of a licensing market for AI training where the threshold question of fair use remains unsettled. The Court declines to award Publishers the extraordinary relief of a preliminary injunction based on legal rights (here, licensing rights) that have not yet been established.” on page 11 of the decision.
