Anthropic's $1.5 billion copyright settlement becomes complicated by author objections
Approval of the $1.5 billion copyright settlement between AI company Anthropic and a group of authors has been delayed after authors raised objections that the proposed lawyers' fees are 'disproportionate'. Federal Judge Vince Chhabria announced at a hearing on Friday that the authors' objections will be considered by the court over the coming 30 days.
The settlement is part of a class action that alleged Anthropic had used authors' copyrighted works without authorisation to train its model. The case, known as Bartz v. Anthropic, is proceeding independently of similar cases pending against ChatGPT-maker OpenAI. About $320 million of the settlement was set to be paid to the law firms running the case.
The $320 million in lawyers' fees proposed by author counsel corresponds to roughly 21% of the settlement total. That ratio is near the upper limit of fee standards accepted by US federal courts in class actions. Some figures from the author group filed with the court arguing the amount has been 'taken unfairly from a historic settlement.'
Four writers who featured prominently in the objection — Ta-Nehisi Coates, Mary Gaitskill, John Banville and Susan Choi — said the principal purpose of the settlement was to protect authors from unauthorised use in AI training, but that the current distribution would leave an average payment of around $1,500 per author. Anthropic, for its part, confirmed that the objections were not aimed at the settlement as a whole.
Anthropic's CEO Dario Amodei said at last week's earnings call that the settlement was important in 'giving the company time to reassess its copyright-compliance infrastructure.' The company said it had set aside the full $1.5 billion in provision in its 2025-26 financial year expenses and that the settlement would have no direct impact on the income statement.
For comparison, the 2008 'Google Books' settlement between Google and authors had faced a similar court-room dispute. At the time objections from within the author group were lodged against the $125 million settlement; the court ultimately rejected the agreement. Judge Chhabria said on Friday that the 'Google Books precedent does not apply directly to this docket, but the underlying principles on the reasonableness of fee structures are shared.'
The structural component of the settlement does not include a commitment by Anthropic not to use authors' works in future model training. The company has instead said that works specifically registered through an 'opt-out' option by authors will not be included in retraining flows. That approach aligns with the inspection mechanism for creative works set out in Article 53 of the EU's AI Act.
A recent US Copyright Office report on AI training and copyright issues encouraged courts 'to develop new interpretation of transformative use between AI companies and creative sectors'. The Anthropic settlement is recorded as the first large case in US federal courts to refer directly to that report.
Observers in Europe say this settlement will also have implications for AI cases on the continent. The 2024 case in Germany between writer Andreas Wendt and creativity platform Aleph Alpha was concluded by a court ruling rather than a settlement; but the Bartz v. Anthropic settlement structure is being used as a reference framework for similar cases in Europe.
After the court reconsiders the Anthropic settlement in the coming 30 days, a decision on whether the settlement will be approved or renegotiated is expected. The author group is preparing an additional proposal to reduce the lawyers' fee ratio to 15% if its court support increases. Anthropic has told the court that it will 'respect any outcome reached by the parties in reasonable agreement.'