US judge approves $1.5B Anthropic copyright settlement, closing a class action for 500,000 works

A US judge has now approved a $1.5 billion settlement in the first big AI copyright class action suit against Anthropic, effectively closing the book on one of the largest cases in US history. The deal, which Anthropic agreed to in September 2025, was approved “this week” as of July 20, and it covers an approximate total of 500,000 works. Under the agreement, affected authors and publishers are to be paid $3,000 for each pirated work.

If you’re looking for the headline number to make sense, here it is in plain terms: $1.5 billion for roughly 500,000 works implies that the court and the settlement framework are pricing the harm at a standardized per-work figure, not an individualized damages fight. That matters because the legal journey for Anthropic was not a simple “win or lose.” In June 2025, a judge ruled that using the plaintiffs' work to train AI models fell under “fair use” within US copyright law. But Anthropic was not legally off the hook anyway, because many of the books used to feed Claude were characterized as “found” online and not lawfully acquired.

So what exactly changed the trajectory from June’s fair use ruling to a $1.5 billion payment approved later? The source points to an uncomfortable nuance that often drives AI litigation: even if training can be argued as transformative, sourcing still matters. The case alleged that Anthropic leveraged a library of millions of “pirated books” to train its Claude chatbot. Discovery of Anthropic’s “found” library originally exposed the company to statutory damages of up to $150,000 for each work. When statutory damages loom, settlement math can shift quickly, even if a court finds part of the conduct could qualify as fair use.

The settlement also arrives with a measure of operational certainty that boards and executives care about. The approximate universe is 500,000 works, and Anthropic says 91% of those eligible have now filed a claim. That claim rate is a signal that the plaintiffs’ side can administer relief, and it reduces the uncertainty that often lingers after major litigation settlements. It also means Anthropic’s exposure is being converted from a long tail of court risk into a more bounded financial obligation with a defined payout mechanism.

It is hard to talk about that number without talking about incentives. Some affected authors argued the payout was too small, and the source highlights why that critique has bite: discovery originally suggested statutory damages could reach up to $150,000 per work. In a world where the upside for tech companies can look enormous, $1.5 billion can feel like a negotiated business cost rather than a calibrated copyright remedy. The source also notes that the private AI company enjoyed a $965 billion valuation back in May. Whether or not you think markets should map to legal outcomes, that valuation backdrop is exactly the kind of context that tends to inflame creators, invite more claims, or influence how future settlements get framed in court filings.

For executives at AI companies, the bigger takeaway is not just the money, it is the message embedded in how courts handled the case. June’s ruling on “fair use” did not erase the risk tied to unlawful provenance. The court’s willingness to approve a massive settlement after that ruling suggests a practical boundary: transformative use arguments may help, but they do not automatically neutralize disputes about whether the training data was lawfully acquired. In other words, a “fair use” win on one legal theory does not guarantee safety if another theory, like unauthorized acquisition, still bites.

And the case is not happening in isolation. The source flags that Google, OpenAI, and Meta are each facing similar court cases over their own AI training practices. When a judge approves a $1.5 billion settlement in a case involving a chatbot trained on alleged pirated books, it becomes a reference point for strategy in other jurisdictions and other lawsuits. Boards will notice that the process did not end with a clean legal ruling that ended the matter. It ended with a settlement that closes a chapter but leaves the broader regulatory and litigation landscape very much alive.

Ultimately, this is a moment of both closure and escalation. For creatives, it is a real payout, at least measured in dollars per work, and a court-approved end to the first big class action. For decision-makers, it is a reminder that AI data pipelines are not just engineering problems. They are compliance, litigation, and capital-allocation problems. The $1.5 billion approval may not be the end of AI copyright battles, but it does reduce uncertainty for Anthropic while setting a concrete benchmark for how similar claims could be quantified, settled, or contested for the rest of the industry.