Anthropic’s $1.5 billion copyright settlement gets final approval

A US federal judge has given final approval to Anthropic’s $1.5 billion settlement with authors and publishers whose books the AI company downloaded from pirate libraries. Judge Araceli Martínez-Olguín signed off on the agreement on July 20, rejecting objections that the amount was too small. Reuters described it as the largest known settlement in a US copyright case and the first major American lawsuit over AI training to end in a settlement.

The case was filed in 2024 by authors including Andrea Bartz, Charles Graeber and Kirk Wallace Johnson. The deal covers more than 480,000 qualifying works. Rights holders are expected to receive roughly $3,000 per work, shared when several parties own rights in the same title.

The legal distinction is important. Former judge William Alsup previously ruled that using books to train Anthropic’s Claude models could qualify as fair use. But he separately found that Anthropic could not build a permanent central library by downloading millions of pirated books from services including Library Genesis and Pirate Library Mirror. Anthropic had also bought and scanned physical books; that route was treated differently. The settlement resolves claims about the pirated copies, not a general claim that all AI training on copyrighted text is unlawful.

Final approval closes the main class action, but it does not create a nationwide rule for the AI industry. The fair-use finding came from one district court and will not be tested on appeal because Anthropic settled the remaining claims. Some authors and publishers opted out and are pursuing separate cases, while Google, Meta, Midjourney, OpenAI and others still face their own copyright disputes. Payment timing may also depend on administration and any appeals.

For creators, the outcome gives data provenance a concrete financial value. For AI labs, it shows that how training material is acquired can create liability independently of what a model later produces. Dataset inventories, licences and a verifiable chain of custody are becoming core engineering and compliance requirements.

Claude users should not expect an immediate product change. The broader effect is on makers and businesses choosing AI suppliers: questions about where training data came from are no longer abstract ethics. They are legal and commercial due diligence. The settlement is neither blanket permission to train nor a nationwide ban; it is a costly warning that acquisition methods matter.