A federal judge in San Francisco delayed final approval of Anthropic’s $1.5 billion settlement with authors alleging unauthorized use of their books to train AI models, citing the need for more transparency on legal fees and payments to lead plaintiffs before signing off on what would be the largest US copyright settlement ever.
- Judge requests detailed breakdown of lawyers’ fees and class leader payments
- Settlement would cover 480,000 works with about $3,000 per work
- Objections include notice issues and allocation of payments per registration
What happened
Anthropic proposed a $1.5 billion settlement in a class-action lawsuit filed by authors who allege the company downloaded more than seven million books from piracy sources to train its AI language models. The settlement covers approximately 480,000 copyrighted works with payments averaging around $3,000 each after deduction of fees.
At a fairness hearing in San Francisco, Judge Araceli Martínez-Olguín refused to finalize the deal without more transparency on how legal fees and lead-plaintiff service awards are calculated. Class counsel had reduced the requested attorney’s fees from 15% to 12.5% of the settlement fund, and proposed payments including about $3 million in expenses and $50,000 each for three lead authors.
Why it matters
This settlement is poised to become the largest copyright-related payout in American history, setting a precedent for how AI companies handle copyrighted training data. The case underscores growing legal challenges as AI development increasingly intersects with intellectual property rights.
The timing is notable as Anthropic is in the midst of fundraising talks targeting a valuation near $900 billion. The $1.5 billion liability represents a significant one-time expense on a balance sheet bolstered by future growth expectations. Court scrutiny of fee distribution echoes broader concerns about class action fairness and transparency.
What to watch next
The court will expect more detailed disclosures on attorney fees, expense reserves, and service awards before deciding whether to approve the settlement. Plaintiffs have registered claims for over 92% of the eligible works, but objections linger regarding adequacy of compensation and whether the notification process was sufficiently broad and accessible.
Future updates could influence how copyright claims are resolved in AI training contexts and may affect investor risk perceptions around AI firms. Monitoring responses from Anthropic, class counsel, and objectors will be critical as this landmark case moves toward final judicial review.