Anthropic just scored a win in the legal jungle that is AI and copyright. A court ruled it was fair game for them to train AI on copyrighted works under the “fair use” doctrine. Translation: they didn’t need to ask nicely or pay up for using those materials, as long as it was transformative. Judge Alsup, a veteran in these waters, labeled the tech as one of the most transformative we’ll see. Sure, we’ve heard that song before, but this time it might hold water.
Chris Mammen, an IP law expert, sees this as a groundbreaking ruling. He notes Alsup’s stance that training an LLM is transformative—even if the AI memorizes content. Alsup isn’t buying the argument that AI training is fundamentally different from human learning. The tech world might want to make some room for this new reality.
This case started as a class action by authors crying foul over Anthropic’s use of their work without a nod or a nickel. While Anthropic claimed a fair use win, Alsup left the door open for authors to pursue claims about pirated content. So, not exactly a clean getaway.
Turns out, Anthropic wasn’t just browsing the library shelves—they downloaded over seven million pirated books. They stopped using these to train AI, but kept them around. Judge Alsup wasn’t impressed, and he’s letting authors seek damages for this little oversight.
Anthropic’s spokesperson sees this as a win for creativity and scientific progress. The plaintiffs’ lawyers, unsurprisingly, had nothing to say. Alsup’s got history with these kinds of cases, having dealt with Google v. Oracle. His experience might heavily influence how future cases play out.
Prior to this, another AI copyright case went the other way. In Thomson Reuters v. Ross, the judge didn’t dig the fair use argument for AI training. That one’s on its way to an appeal. With Alsup’s decision, the fair use line might get clearer, especially when piracy isn’t in the mix.
While some are waving the victory flag for fair use, Alsup was clear: piracy is a no-go. He detailed Anthropic’s strategy of downloading pirated content from databases like Books3, LibGen, and PiLiMi. Not exactly a masterclass in copyright respect.
Anthropic isn’t alone in the hot seat. Meta’s also facing fire for allegedly using pirated books in its AI training. The minimum penalty for this kind of infringement is $750 per book, and with Anthropic’s stash hitting seven million, the numbers are daunting. No trial date yet, but this saga is far from over.