Anthropic, the company behind the Claude AI model, has agreed to a $1.5 billion class-action settlement after being caught using pirated books to train its flagship product. The case, known as Bartz v. Anthropic, covers approximately 500,000 copyrighted works and will pay qualifying US-registered rightsholders roughly $3,000 per title.
The Irish Writers’ Union has now entered the picture, reaching its own arrangement with Anthropic and actively advising Irish authors on how to file claims. The union sees the case as a potential precedent for European copyright enforcement against AI companies.
What actually happened
The lawsuit alleged that Anthropic downloaded books from Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi), two well-known piracy platforms, to feed its language models during training.
A US District Court issued a mixed ruling that drew a surprisingly clear line. Training AI on legally acquired material? That could qualify as fair use. Downloading copyrighted works from pirate sites to build your dataset? That’s plain infringement.
As part of the settlement, Anthropic is required to destroy the pirated datasets used in training. Final court approval is expected around mid-2026. Payments will be distributed in installments after necessary deductions.
Why the Irish Writers’ Union matters here
The IWU’s involvement signals that this isn’t going to stay an American story. The union is positioning itself as a guide for European authors who may have claims, and its public support for the settlement framework suggests organized labor in the creative industries sees this as a template.
Europe already has stricter data protection and copyright regimes than the US. The EU’s AI Act, which began phasing in during 2024, requires transparency about training data.
The AI industry’s growing copyright problem
Anthropic is far from the only AI company facing these questions. OpenAI has been sued by the New York Times and multiple authors. Stability AI has dealt with lawsuits from Getty Images.
The $1.5 billion settlement number sounds enormous, but Anthropic was recently valued in the tens of billions of dollars. If every AI company needs to either license its training data or risk billion-dollar settlements, the economics of building large language models change fundamentally.