US technology companies, like Mark Zuckerberg said of Facebook over a decade ago, are now used to moving fast, as well as breaking things. One thing that has certainly been sacrificed on the altar of progress is copyright in respect to AI, and how the largest firms have built at least part of their models on the IP of people.

There has been some early guidance from the US District Court for the Northern District of California, which ruled in favour of Meta and Anthropic in two cases over the use of copyrighted material. In Bartz vs Anthropic, the case concerned Anthropic acquiring millions of books through purchase, scanning, as well as downloading pirated copies, and then using them to train its Claude model. In this case, the court said that Anthropic’s use of legally acquired books to train its model was fair use, but this was not the case where it had trained its model on a library of pirated books. This could cause harm to the market, and there could be statutory damages down the line for wilful infringement.