A US judge has indicated that he will cut back a lawsuit filed against Meta by three writers - including comedian Sarah Silverman - in relation to the training of the tech giant’s generative AI platform LLaMA with copyright protected works, including those written by the plaintiffs.
Californian judge Vince Chhabria was considering Meta's motion to dismiss entirely most of claims made by the authors in their lawsuit, which was filed in July.
According to Law360, Chhabria told a legal rep for Silverman et al that he understood the core copyright infringement claim in his clients' lawsuit - ie that Meta allegedly made copies of their books as part of the training process without licence. Which looks like copyright infringement.
Meta - like most AI companies - will claim that using copyright protected works in that way constitutes 'fair use' under US law, so a licence is not required from the copyright owner. But that's an argument to be considered at another time.
However, the authors' lawsuit also makes various other claims of copyright infringement against Meta in relation to the development and use of LLaMA, and also accuses the tech firm of breaching rules in the US Digital Millennium Copyright Act around copyright management information.
And on those other claims, Chhabria is less convinced. Indeed, he said in court this week: "I understand your core theory of copyright infringement ... Meta used the books to train its program ... [but] your remaining theories of liability, I don't understand, even a little bit".
In a subsequent discussion about those other claims, the judge implied that - while he is likely to dismiss them from the lawsuit in their current form - he might accept revised and much more narrowly defined versions of said claims if submitted in an amended complaint.
Although this is not a music case, all the ongoing litigation around the copyright obligations of generative AI platforms are relevant to music companies. Because they should provide some clarity about the extent to which AI companies must secure licences from the music industry when making use of existing recordings and songs in order to train their models.