This is just one of many lawsuits against AI companies claiming they copied and used copyrighted materials in training their models, that hinges on the fair use provision of the copyright laws
This resurfaced overnight after Chatgpt’s new image generator was used en masse to produce images in the style of Studio Ghibli, which had obviously been used in training the model in so far that it could easily mimic their style.
Music industry denied
In the Anthropic case, publishers had sought a preliminary injunction against Anthropic for causing a whole list of harms to both their reputation and the market value of their lyrics, according to TorrentFreak.
Anthropic had responded with the classic AI response, claiming that using publicly available data in training their models is fair use, not reckless copying. Both OpenAI and Meta have used this defense in similar lawsuits.
As for the specific lyrics claimed in the lawsuit, Anthropic went as far as to say that the lyrics were indeed included in their training data, but «guardrails» had been put in place to prevent them from being published verbatim by their chatbot.
California District Court Judge Eumi K. Lee, found no irreparable harm that the publishers had suffered to be mitigated by an injunction to remove the lyrics from the database.
Ever expanding list of songs
The inclusion of even unwritten songs in the damages, would make it impossible to Anthropic to clean up, had they prevailed.
— The enormous and seemingly ever-expanding scope of Works included in the requested injunction raises significant concerns regarding enforceability and manageability, wrote the judge.
The judge finally ruled on a motion to dismiss the case from Anthropic, and granted the request on all points, but left out the direct copyright infringement claim.
That should keep the court case alive for another round, giving the music publishers 30 days to amend their complaint in accordance with the courts rulings.
The case continues
Anthropic was happy with the preliminary ruling, that seemed to go mostly in their favor, and hinted at the broader defence as the case continues:
– As the case continues, we look forward to explaining why use of copyrighted material for training large language models aligns with fair use principles under copyright law, they say in a statement for Gizmodo.
In response to this «setback,» lawyers from the publishers they are still optimistic on the broader case, and says the guardrails Anthropic put in place were proof that they had indeed used the copyrighted lyrics in their training, telling Music Business Worldwide:
— Anthropic had already conceded the merits of our claims against its infringing outputs of our copyrighted song lyrics, by entering into a stipulation requiring it to maintain ‘guardrails’ to prevent such infringing outputs, thereby resolving a critical aspect of the motion in our favor.
While there are many pending court cases revolving on this precise premise — if AI training data infringes on the copyright of the rightsholders — none of them har been brought to fruition.
This has been described as a sword hanging over AI companies, because if they should lose and see a precent be set, they’ll have to refigure the whole training process and rely only on non-copyrighted materials, severely diminishing the usefulness of AI models.
Read more: Play by play coverage from TorrentFreak, writeup from Gizmodo, and quotes from Music Business Worldwide.