Concord Music Group et al. v. Anthropic PBC was a case in the Northern District of California involving another artificial intelligence company defending itself against hypothetical infringement.
Anthropic is a technology company whose main product is an artificial intelligence called Claude. Anthropic trains Claude by creating a training corpus, which consists of large amounts of content from the internet. Anthropic then creates a training data set which fine-tunes Claude through reinforcement learning based on an AI feedback, and then the models generate output to the consumers.
The plaintiffs are publishers of music who brought suit for copyright infringement, vicarious infringement, and intentional removal of copyright management information (CMI). They claimed that Claude was engaging in the wholesale copying of copyrighted lyrics.
Anthropic filed a motion to dismiss all the claims except for the direct copyright infringement claims. The court addressed the motion to dismiss the infringement, vicarious infringement, and intentional removal of CMI (content management information).
The court granted the motion to dismiss those three claims.
For the vicarious infringement claim, the publishers had to allege that Anthropic: 1) had knowledge of infringement; and 2) materially contributed or induced that infringing conduct by others. According to the plaintiffs, Anthropic’s business customers might use the Claude tool to seek copyrighted lyrics. The court said that the fact that Anthropic had the ability to do something was not sufficient to cover the predicate act necessary for direct third-party infringement.
The publishers did not sufficiently allege that Anthropic knew about specific acts of third-party infringement, or even had reason to know. Plaintiffs only contended that Anthropic was well aware of its licensees’ and users’ infringing conduct because it knowingly trained its AI models through infringing conduct – the unauthorized copying of lyrics.
The court rejected those claims, calling them conclusory. It was possible those lyrics could be used for lawful as well as unlawful uses. One cannot get a conclusion of liability for a third party’s infringing conduct when the technology can provide both lawful and unlawful uses. That’s not sufficient to satisfy the pleading standard for vicarious infringement.
For vicarious infringement, a plaintiff has to show that the defendant has the ability to supervise the infringing conduct, and also has a direct financial interest in the infringing activity. However, the publishers didn’t allege enough to show that there was direct infringement by Anthropic users. As such, the court granted the motion to dismiss with leave to amend as to the count for vicarious infringement.
CMI is information like the title, the copyright owner and the terms and uses of the work. Provisions in the Digital Millennium Copyright Act (DMCA) forbid removal of such CMI, but require the defendant to possess the state of knowing, or a reasonable basis to know, that removing the CMI would induce, enable, facilitate, or conceal infringement.
In this case, the publishers merely repeated the statutory language without stating any facts to show that Anthropic intentionally removed CMI. The allegations were too conclusory to establish that Anthropic knew of any removal of CMI, if there was any such removal. That claim was also dismissed with leave to amend the complaint. Further burying this claim, some of Claude’s outputs actually included attribution for some lyrics — undermining the theory that all CMI had been removed, as plaintiffs had alleged.
A question that the court would leave unanswered was the propriety of the plaintiffs’ data collection. The publishers had attempted to obtain verbatim copies of lyrics through the use of investigators who expressly asked for lyrics that the publishers owned. The court declined to decide whether plaintiff’s use of investigators was sufficient to be the predicate act required for third-party infringement. After this motion to dismiss was granted and the plaintiffs would have a chance to amend their complaint.
It’s important to note that despite all of the angles by which traditional copyright concepts are being used to try and prevent artificial intelligence from marginalizing content holders’ rights, content holders may just be poorly served by existing copyright laws.
