Authors object to proposed settlement involving Anthropic
A group of authors has formally rejected a proposed class action settlement involving Anthropic, arguing that the deal would allow large language model companies to resolve sweeping claims too cheaply and too quickly. In their objection, the authors said “LLM companies should not be able to so easily extinguish thousands upon thousands of high-value claims at bargain-basement rates,” framing the settlement as a test of how courts will value alleged harms tied to the use of copyrighted works in AI training.
The dispute sits at the intersection of copyright law, emerging AI business models, and the growing wave of litigation targeting how generative AI systems are built. While settlement terms were not detailed in the authors’ statement, their criticism signals a broader concern among creators: that class action deals could set precedents that limit future recovery, reduce leverage for individual rights holders, or normalize compensation levels they view as inadequate.
Why the objection matters for LLM litigation
Class action settlements can have far-reaching effects beyond a single company. If approved by a court, they may bind large groups of potential claimants—sometimes including people who never actively participated—depending on how the class is defined and how opt-out provisions are structured. That is why objections like this can become pivotal: they can prompt judges to demand stronger justification for the settlement amount, clearer notice to potential class members, or changes to how claims are released.
The authors’ language suggests they believe the claims at issue are substantial, both in number and in value. Their reference to “thousands upon thousands” of claims indicates concerns that a broad release could cover a wide set of works and alleged uses, potentially limiting separate lawsuits or individual negotiations. In AI-related cases, the central allegation often involves the use of copyrighted books, articles, or other materials during training or fine-tuning, raising questions about consent, attribution, and compensation.
Key tension: speed and certainty vs. valuation
Settlements are often attractive to defendants because they reduce uncertainty, cap liability, and avoid prolonged discovery. Plaintiffs may accept them to secure guaranteed compensation and avoid years of litigation. But the authors’ objection points to a fundamental tension: they argue that the deal’s price, not merely its structure, is out of step with the scale of alleged infringement or harm.
That criticism also reflects a strategic concern. If multiple AI companies face similar lawsuits, a low-dollar settlement in one case may influence negotiation expectations in others, even if it is not legally binding precedent. For rights holders, that can feel like a market signal that undervalues creative work in the context of AI development.
Broader implications for the AI industry
For companies building generative AI, legal exposure related to training data has become a significant business risk. The cost of litigation, potential damages, and the operational impact of changing data practices can affect product roadmaps and investor sentiment. As AI models grow more capable and more widely deployed, scrutiny has intensified around the provenance of training data and whether existing legal frameworks adequately address large-scale machine learning.
The authors’ objection underscores the possibility that some creators will resist class-wide resolutions in favor of individualized claims, higher compensation, or more stringent restrictions on how their works can be used. That could complicate efforts by AI firms to “clear the deck” through broad settlements, particularly if courts become skeptical of deals that release expansive rights without what objectors view as meaningful consideration.
Potential outcomes if the court engages the objection
When a settlement faces serious objections, courts can take several paths. A judge may:
- Request additional information to assess whether the settlement is fair, reasonable, and adequate for the class.
- Require changes to notice procedures so more affected authors understand their rights and options.
- Encourage renegotiation of monetary terms or carve-outs that preserve certain claims.
- Reject the settlement outright if it fails to meet legal standards.
Even without a final ruling, the objection can influence the negotiation dynamics between plaintiffs’ counsel and the defendant, potentially leading to improved terms or a narrower release of claims.
Creators push back on “bargain-basement” resolutions
The authors’ statement is notable for its blunt characterization of the proposed deal as a “bargain-basement” outcome. That framing suggests they see the case as more than a dispute over a single settlement figure. Instead, they appear to be arguing for a principle: that the value of creative works and the scale of alleged unauthorized use should be reflected in any resolution involving LLM development.
As generative AI becomes embedded in consumer products and enterprise tools, the question of how creators are compensated—or whether permission is required at all—continues to be debated in courts, legislatures, and industry negotiations. Objections like this one signal that some authors are prepared to contest not just liability, but also the mechanisms by which AI companies seek to close out claims.
What to watch next
The next steps will depend on how the court evaluates the settlement and the merits of the objection. Observers will be watching for indications of how judges weigh the complexity of LLM training practices against traditional copyright principles, and how they assess the fairness of class action deals in cases involving large numbers of works and claimants.
For Anthropic and other AI developers, the proceedings may offer a clearer picture of the legal and financial contours of resolving training-data disputes—either through settlements that withstand scrutiny or through continued litigation that tests the boundaries of copyright law in the age of generative AI.
Related
