A federal judge has put the brakes on Anthropic's $1.5 billion copyright settlement, declining to give it immediate approval and adding a new layer of legal uncertainty to one of the largest AI training-data cases on record.
The settlement, reached with authors and publishers who sued Anthropic over its use of copyrighted books to train Claude, requires court sign-off before it can take effect. Judges reviewing class action settlements routinely scrutinize them for fairness to all class members - and a $1.5B payout can attract objections from parties who feel the terms undervalue their claims or improperly release future rights.
The delay doesn't kill the deal, but it does extend a legal overhang that the AI industry has been watching closely. Courts approving or blocking these settlements will effectively set the price of training data retroactively - and signal to every AI company what liability looks like when you build products on copyrighted text without licensing agreements.
For Anthropic, the timing is awkward. The company raised $2.5 billion from Google and others at a $61.5 billion valuation earlier this year and is competing directly with OpenAI on frontier model development. A drawn-out legal process keeps attention on how Claude was trained rather than what it can do.
The broader copyright fight against AI companies isn't going away. OpenAI, Meta, and others face parallel suits from authors, musicians, and news organizations. How Anthropic's settlement ultimately resolves - including what rights holders actually receive and what future training practices must look like - will influence how those cases are argued and settled.