A US federal judge has delayed final approvals of Anthropic’s landmark $1.5 billion copyright settlement over its widespread use of books to train artificial intelligence models, following objections raised by several authors and class members.
On Thursday, US District Judge Araceli Martinez-Olguin declined to rubber-stamp the settlement, which is considered the largest copyright settlement in US history. Instead, she sought to better understand the rationale of objecting class members who chose to opt out. She requested that the lead plaintiffs address the key concerns of objectors, who argued that the legal team's compensation was excessively high while payouts to the actual class members were a 'pittance.'
Objections reviewed by Ars Technica reveal allegations that the authors' legal team attempted to unfairly shut out dissenting voices. Objectors noted that the lawyers requested over $320 million in legal fees, while individual authors are only expected to receive approximately $3,000 each. They have urged the court to delay approval until a fairer compensation plan is formulated.
'Every dollar that Counsel takes from the Settlement fund is one that is not given to those actually harmed,' wrote Pierce Story, an objector and author of two works included in the class. Story calculated that the requested payout equates to an hourly rate of $10,000 to $12,000 for the lawyers, which he deemed excessive, citing precedent from a T-Mobile case where a court noted that 'no reasonable class member would willingly pay' even a lower requested fee of $7,000–$9,500 per hour.
Story further accused the lawyers of reneging on a promise to tie their compensation to actual member payouts. He expressed frustration that the requested legal fees are tied to the entire $1.5 billion fund, even though many eligible authors have yet to register and may never receive compensation.
While an attorney for the authors confirmed that claims have been filed for over 92% of the 480,000 works covered by the settlement, objectors maintain that attorney fees should scale with the number of actual claimants, rather than the total fund. Story suggested that capping attorney fees at a still-generous $70 million would drastically increase the compensation distributed to the authors who suffered the actual harm.
[AgentUpdate Depth Analysis] The roadblock in Anthropic’s $1.5 billion settlement underscores the compounding legal liabilities of 'train now, pay later' approaches in generative AI. While competitors like OpenAI lean heavily into proactive licensing deals with major publishers, Anthropic’s legal entanglement signals a shifting paradigm where data provenance dictates model viability. For the broader AI Agent ecosystem, this is a critical wake-up call. Agent architectures rely on orchestrating complex workflows powered by foundational LLMs. If the underlying models are legally compromised, the entire upstream Agent ecosystem risks disruption. Moving forward, high-quality, legally bulletproof training corpora will become a premium moat. Developers must scrutinize the data compliance of their model providers, as the future of sovereign AI Agents will be built on legitimate intellectual property rather than scraped grey-area datasets.
