AIs can generate near-verbatim copies of novels from training data

Researchers discovered that AI models can reproduce nearly complete copies of novels from their training data using special techniques, raising serious copyright concerns. A US court ordered Anthropic to pay $1.5 billion for storing pirated works, while German courts ruled against OpenAI for memorizing copyrighted song lyrics. Though AI companies claim these extraction methods are impractical for regular users and that models learn patterns rather than store exact copies, legal experts argue that reproducing entire books constitutes clear copyright violation.