- Dotika
- Posts
- Anthropic pays $1.5 billion for pirated books
Anthropic pays $1.5 billion for pirated books
ALSO : OpenAI's massive AI spending forecast
Geoffrey NICHIL
September 08, 2025
News of the day
1. Anthropic settles $1.5 billion lawsuit for using pirated books to train AI. This landmark case sets a precedent for copyright recovery in the AI era, covering 500,000 books and highlighting ethical data sourcing concerns.→ Read more
2. OpenAI triples its 2029 spending forecast to $115B, significantly increasing its cash burn estimate. Revenue is also expected to rise. → Read more
3. Amazon Music's new 'Weekly Vibe' feature uses AI to create personalized playlists every Monday, adapting to users' evolving musical moods and interests for fresh discovery. → Read more
4. Grok 4.2, codenamed Sonoma Sky, is poised to be a significant advancement in LLMs, showing exceptional performance on benchmarks and impressive steerability. Its large context window and efficiency are key highlights. → Read more
Our take
Hi Dotikers!
Anthropic has agreed to pay at least $1.5 billion to settle a class action brought by authors who accused it of using pirated books to train Claude. The settlement, which is subject to court approval, provides for the destruction of the illicit files. It sets compensation at around $3,000 per work for nearly 500,000 titles, and is presented by the plaintiffs as the largest recovery ever obtained in a copyright case. Anthropic does not admit wrongdoing.
This outcome follows a key decision handed down in June by Judge William Alsup. The court held that training models on books falls under fair use and is therefore lawful, while finding unlawful Anthropic’s creation of a centralized library of more than seven million pirated books. The trial scheduled for December is thus averted, but the debate over the scope of fair use remains open, as other cases also target OpenAI, Microsoft, and Meta.
Ironically, the case breaks the same week Anthropic announces a $13 billion raise, bringing its valuation to $183 billion. Financial strength doesn’t erase legal costs; it makes them visible.
The era of permissionless scraping is coming to an end. The winners will be those who buy rights, document the provenance of their data, and know how to purge contaminated corpora. Second, compliance becomes a strategic budget line alongside GPUs, with painful trade-offs between deployment speed and legal safety. Finally, this settlement doesn’t close the matter; it sets a reference price that will weigh on upcoming negotiations with publishers and rightsholders. The AI Wild West has just installed a tollbooth ; better to arrive with a pass than with a mask 😉
G.
Meme of the day
Reply