MBW Explains is a series of analytical features in which we explore the context behind major music industry talking points – and suggest what might happen next. Only MBW+ subscribers have unlimited access to these articles. MBW Explains is supported by Reservoir.
Universal Music Publishing Group, Concord Music Group, and ABKCO filed a second copyright infringement lawsuit against AI giant Anthropic last week, and the complaint goes significantly further than the first.
The 48-page filing, submitted in the US District Court for the Northern District of California, doesn’t just target Anthropic PBC. It also names CEO Dario Amodei and co-founder Benjamin Mann as individual defendants, alleging they personally participated in and directed infringing activity.
The decision to name founders personally is a significant escalation in music rightsholder litigation against an AI company.
The numbers cited in the complaint are also staggering. As MBW reported last week, the lawsuit covers 714 works related to alleged torrenting activity and 20,517 works related to AI training and output claims.
With statutory damages of up to $150,000 per work available, potential liability could exceed $3 billion. You can read the new filing in full here.
Beyond the headline figures, however, the complaint contains several legal arguments that could have far-reaching implications for how AI companies source their training data – and how courts evaluate those practices.
“We believe this will be one of the largest (if not the single-largest) non-class action copyright cases filed in the US.”
UMG, Concord, et al
Last week, the plaintiffs told MBW in a statement that they “believe this will be one of the largest (if not the single-largest) non-class action copyright cases” filed in the United States.
“We have been compelled to file this second lawsuit against Anthropic because of its persistent and brazen infringement of our songwriters’ copyrighted compositions taken from notorious pirate sites”, they said.
“The new case also addresses Anthropic’s ongoing violation of these rights by exploiting lyrics in the training of new AI models without authorization, as well as in the outputs generated.
“In total, we are suing for infringement of more than 20,000 songs, with potential statutory damages of more than $3 billion.”
As the case gears up to be one of the most significant AI copyright battles yet, here are six key arguments from the publishers’ complaint…
1. ‘Torrenting was a standalone act of unmistakable, irredeemable infringement’
The publishers argue that Anthropic’s alleged use of BitTorrent to download books from pirate libraries constitutes copyright infringement regardless of what Anthropic subsequently did with those files.
The complaint states: “Defendants unlawfully torrented Publishers’ works to amass a vast central library of written texts Anthropic would maintain forever.
“To the extent Defendants now try to absolve themselves of liability for this blatant theft by claiming that Anthropic later used some subset of these stolen works for AI training, any such claimed use is irrelevant (and would not in any case qualify as fair use). Defendants’ piracy of each of Publishers’ musical compositions via torrenting was a standalone act of unmistakable, irredeemable infringement.”
In other words, the publishers argue that even if Anthropic claims a fair use defense for AI training, that defense cannot retroactively justify the initial act of downloading pirated copies from illegal websites.
The complaint adds: “Regardless of Anthropic’s later use, its piracy of these books via BitTorrent was unquestionably infringing. Even if some subset of the books Defendants illegally torrented were sometimes used for AI training, that cannot excuse their mass torrenting of millions of pirated books without paying for them.”
2. ‘BitTorrent means Anthropic was simultaneously uploading infringing copies’
The publishers make a technical argument about how BitTorrent works that could significantly expand Anthropic’s potential liability.
Unlike traditional downloading, BitTorrent operates on a peer-to-peer basis, where users simultaneously upload files to others while downloading. The complaint explains: “Once a user downloads a piece of a file, the user immediately becomes a distributor of the file to others, creating a ‘swarm’ where everyone downloading the file also acts as a distributor of the file to others.”
The publishers argue this means Anthropic didn’t just reproduce their works – it also distributed them to the public, violating a separate exclusive right under copyright law.
The complaint states: “When Defendants downloaded copies of these pirated books via torrenting, they violated Publishers’ exclusive right of reproduction in these works. And to make matters worse, because of the two-way nature of the BitTorrent protocol, when Defendants downloaded copies of these pirated books via torrenting, they simultaneously uploaded to the public unauthorized copies of the same books, thereby infringing Publishers’ exclusive right of distribution in these works and contributing to further infringement of Publishers’ works as well.”
The filing adds: “Each pirated work Defendants torrented was likely shared thousands if not tens of thousands of times, depriving Publishers of substantial revenue.”
3. ‘Anthropic created a permanent central library – regardless of AI training’
A key element of the publishers’ argument is that Anthropic allegedly copied these works not merely for AI training, but to build a permanent repository of text for multiple purposes.
The complaint states: “Defendants copied these books, including those containing Publishers’ works, via torrenting in order to amass a vast, general-purpose central library of copyrighted works and other written text that Anthropic could keep forever and use for whatever purpose it wished.”
The publishers allege that Anthropic maintained copies of torrented files “in the same format as they had originally torrented them” and that “regardless of whether specific text was utilized for AI training or not, Anthropic maintained unlawful copies of the text as part of its central library, with the goal of storing these copies ‘forever.'”
The complaint argues this undermines any potential fair use defense: “That copying is an undisputed act of infringement with no plausible defense of any kind.”
4. ‘Founders personally directed and controlled the illegal torrenting’
By naming Dario Amodei and Benjamin Mann as individual defendants, the publishers are attempting to pierce the corporate veil and hold Anthropic’s leadership personally liable.
The complaint alleges that before founding Anthropic, Amodei and Mann “led OpenAI’s effort between 2019 and 2020 to torrent books from known collections of pirated books available on illegal websites.”
Regarding their activities at Anthropic, the filing states: “Dr. Amodei and Mr. Mann were primary participants and moving forces behind this illegal torrenting of millions of books, including Publishers’ works, from LibGen and PiLiMi by Defendants.
“At Dr. Amodei’s direction and with Dr. Amodei’s express approval, Mr. Mann personally engaged in the illegal torrenting, and both Dr. Amodei and Mr. Mann personally directed and controlled this torrenting activity.”
The complaint also alleges that Amodei acknowledged internal concerns about the legality of using LibGen, but proceeded anyway: “Dr. Amodei himself had described LibGen as ‘sketchy.’ Yet Dr. Amodei and others approved the torrenting.”
5. ‘Anthropic deliberately removed copyright information it deemed useless junk.’
The publishers bring a separate claim under Section 1202 of the Copyright Act, which prohibits the removal or alteration of “Copyright Management Information” – data identifying copyrighted works and their owners.
The complaint alleges that Anthropic deliberately used extraction tools to strip copyright notices from training data, citing internal communications in which employees compared tools based on their effectiveness at removing such notices.
According to the filing, when one tool left copyright notices intact, Anthropic employees considered it inferior: “In one chat, a member of Anthropic’s technical staff shared an example of jusText’s purported deficiencies with Mr. Mann and Dr. Kaplan: when applied to a scraped webpage containing footnotes, a copyright owner name, and ‘© 2019’ copyright notice, jusText left that information untouched. In contrast, Newspaper, which removed the footnotes, copyright owner name, and copyright notice entirely, was considered ‘a significant improvement.'”
The complaint states: “Because Newspaper removed Copyright Management Information more effectively, Anthropic purposefully decided to employ that tool to remove copyright notices and other Copyright Management Information from Publishers’ lyrics and other copyrighted works. In making that decision, Anthropic dismissed this critical information as ‘useless junk’ to be scrubbed from Claude’s training dataset.”
6. ‘Anthropic’s guardrails are a band-aid, not a cure’
The publishers acknowledge that Anthropic implemented guardrails following the first lawsuit to limit infringing output, but argue these measures are insufficient and don’t address the underlying infringement.
The complaint states: “Although Anthropic’s guardrails may have addressed some of the most egregiously infringing output that Claude frequently generated prior to Concord I, they still do not prevent a wide range of prompts and outputs implicating Publishers’ lyrics and violating Publishers’ rights.”
The filing alleges that Anthropic “deliberately chose to include lyrics for only a limited number of specific songs as part its guardrails (including the 500 Works in Suit identified in Concord I), such that those guardrails will not comprehensively prevent output copying lyrics from the much broader universe of copyrighted songs beyond that limited set chosen by Anthropic.”
The publishers argue: “What’s more, because these guardrails address only Claude output, and do nothing to prevent Anthropic’s underlying exploitation of Publishers’ lyrics in AI training, they are at most a band-aid—not a cure—for Anthropic’s infringement.”
Music Business Worldwide
