30-06-2025
AI Gets A Pass On Copyright As Judge Favors ‘Transformative Tech'
Judge rules Anthropic can train AI on books it bought and scanned — so long as it destroys ... More originals.
Anthropic can continue to train its generative AI models on copyrighted books, as long as it scans legally obtained books into a computer and destroys the originals. Judge William Alsup sided with Anthropic on Monday, calling generative AI 'the most transformative [technology]The decision dismissed most claims in the lawsuit filed by authors who alleged that their books were used, without permission, to train Anthropic's Claude AI models. Only one claim — concerning Anthropic's initial use of pirated copies for its central library — was allowed to proceed. The rest were wiped off the board. The company will still need to resolve, through a trial, its training practices using more than seven million pirated titles.
Overall, it's still viewed as a win for Anthropic. But it comes at the cost of clarity in copyright law.
Expert Reaction To AI Copyright Decision — 'This is bad law'
Yelena Ambartsumian doesn't mince words.
'If you've read the Bartz v. Anthropic ruling on copyright infringement, you'll see the order is bad law,' said Ambartsumian, IP attorney and founder of Ambart Law, PLLC. 'It's clear Judge Alsup wanted to protect and promote generative AI development and that's exactly what he did.'
Her take? The ruling ignores core copyright principles. Most glaring, she said, is the judge's logic around reproductions. The Copyright Act gives authors the exclusive right to make copies of their work. That includes the kind of large-scale duplication used during AI training. But Judge Alsup wrote that no infringement occurs 'where one copy entirely replaced the other.'
Ambartsumian is blunt. 'According to the decision, if I scan a book to make a digital copy, and then destroy the original book, there is no copyright infringement, because my digital copy replaced it. That can't be.'
Fair Use For AI Gets Contorted
Judge Alsup also leaned heavily on fair use. But the analysis was thin where it mattered most.
Instead of focusing on the commercial use of protected works, the ruling compared AI training to how people learn. That's a slippery slope since humans don't monetize their thought patterns at industrial scale.
'This is a commercial machine, not a human learner,' Ambartsumian said. She acknowledges the comparison might hold conceptually, but that doesn't mean the legal treatment should be the same. 'I've written before that I think copyright law may not be the best avenue to address issues of un-remunerated copying,' she added. 'But this reasoning still skips the most important parts of the fair use test and does not analyze the commercial aspects at play.'
She added that Judge Alsup also disregarded the scope of what was taken. Anthropic didn't just borrow snippets. It ingested entire books. But instead of addressing that head-on, the court focused on whether Claude needed to use full books or just a subset. The conclusion? Using everything available makes the model better, so it's allowed.
'That's exactly what I warned about,' Ambartsumian said, referencing her reaction to the Copyright Office's earlier report on AI training. 'The Copyright Office posited that training on massive amounts of diverse works could make a model more robust and thus more likely to be 'transformative.' This invites more copying, not less, and could incentivize copyright infringement.'
Judicial Discretion And Selective AI Model Sympathy
There's another current running under the surface. Anthropic may have benefited from being seen as the 'cleaner player' in the AI race.
'Would Anthropic's arguably-less-ethical older brother, OpenAI, have gotten the same result from the same judge? Probably not,' Ambartsumian said. 'In copyright law, we often see judges twisting case law to incentivize 'good guys' and punish 'bad guys' – or ignore creators.'
That inconsistency now bleeds into courtrooms. In the ongoing New York Times lawsuit against OpenAI, the tone from the bench has been colder, the skepticism sharper.
So where does that leave everyone? In legal limbo. The provisional appeal window remains open. But Ambartsumian doesn't expect much.
'As a litigator and copyright-law-obsessive, I'm disappointed,' she said. 'I would have liked to see a more robust analysis of why Anthropic's training was transformative. But as a legal realist, I don't think an appellate court will reach a different outcome.'
What began as a fight over rights ended as a referendum on priorities. The court chose innovation. Creators were left with the scraps.
