Image: © photo for everything/Stock.adobe.com
Anthropic argued that excluding an ‘undefined amount of unknown material’ from its training corpus would be ‘virtually impossible’.
Yesterday (25 March), a California judge sided with artificial intelligence (AI) start-up Anthropic by denying a motion for injunction filed by mega music publishers that would have stopped the start-up from using their song lyrics to train its AI models.
The copyright lawsuit was first filed in October 2023 by Universal Music Group, Concord and Capitol CMG, among several other large music publishers, which collectively control the rights to millions of songs.
The initial complaint alleged that Anthropic “unlawfully copies and disseminates vast amounts of copyrights works”, including song lyrics, in the process of building and operating its AI models.
Requesting the motion for injunction, the plaintiffs clarified that they “are not asking Anthropic to have to retrain existing models” or “to pull models out of the marketplace”. Moreover, the request would not include “any models that are currently under development”, they added.
However, US district judge Eumi Lee found that even though the publishers have tried to clarify the scope of the proposed injunction, the details remain “poorly defined”. The complaint references 500 songs, however at the hearing, the plaintiffs confirmed that the injunction would extend to “all of the publishers’ works”, the order read.
While Anthropic argued that excluding an “undefined amount of unknown material” from its training corpus would be “virtually impossible”.
Moreover, Lee further wrote that the publishers were neither able to prove reputational harm from Anthropic’s alleged actions, nor were they able to demonstrate how Claude’s usage of their work is diminishing their value in the market.
Rejecting the motion yesterday, Lee said that an injunction requiring Anthropic to retrain already-released AI models or to rebuild the corpus it uses to train its models “could impose unforeseeable costs” on the company.
AI and copyright holders have been duelling for a number of years now. The unprecedented growth at which generative AI and large language models have grown has left lawmakers struggling to catch up.
In August 2024, Meta and Universal Music Group struck a deal that saw the Facebook parent start cracking down on unauthorised AI-generated content that violates the copyright of the group’s musicians.
While recently, in response to proposed changes to UK copyright law that would allow AI developers to train their models using copyrighted material, more than 1,000 artists, including Annie Lennox, Kate Bush, Billy Ocean and Hans Zimmer, released a silent album titled ‘Is This What We Want?’.
Similar lawsuits are being filed and fought. Last month, more than a dozen top news publishers, including Forbes, Condé Nast and Politico filed a joint lawsuit against Canadian AI company Cohere over allegations of “systematic copyright and trademark infringement”.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
