This court decision could affect future licensing fees more than infringement payouts.
SOPA Images/LightRocket via Getty Images
Anthropic won’t have to delete a single training parameter or scrub the memory of its leading large language model Claude.
That’s the headline buried inside the proposed $1.5 billion settlement Anthropic just reached to resolve one of the largest copyright class actions in U.S. history. A group of authors filed a class action suit against the AI firm, alleging mass infringement through pirated digital library files. The case was headed for trial — until Friday’s decision.
“This $1.5 billion sum, for about half a million copyrighted works, amounts to roughly $3,000 per pirated work,” said Yelena Ambartsumian, an AI governance and IP lawyer and founder of Ambart Law PLLC. “It sounds like a lot of money, and it is, until you remember plaintiffs could recover $150,000 statutory damages per work for willful infringement.”
The math pencils out — Anthropic got off cheap.
No Deletion Or AI Model Disgorgement — Just A Receipt
A critical carveout of the proposed agreement is that Anthropic doesn’t have to delete or modify its Claude models. Instead, the company will only destroy the infringing digital library it used to train those AI models.
So basically, the $1.5 billion payout will serve as a retroactive licensing fee, not an actual penalty. There’s no clawback clause. No model overhaul required. No AI system shutdown.
“This isn’t algorithmic disgorgement,” Ambartsumian said. “In copyright law, one of the remedies is the destruction of the infringing work. But that’s not happening here. Because Anthropic gets to keep its models, this settlement can be viewed as a licensing fee.”
That framing changes everything. It opens the door for other LLM makers to treat copyright lawsuits as just another line item. This creates a train now, pay later business model for AI firms.
While AI Class Status Mattered — Timing Matters More
It’s worth noting that the court certified a class of rights holders, which won’t be the case in every copyright claim against AI model developers.
“Not all the lawsuits against OpenAI, Midjourney, Google and others are class actions,” Ambartsumian noted. “Class certification is a big deal. Those with valid claims will still bring them, but this ruling doesn’t necessarily trigger a flood of new lawsuits.”
Judge Alsup’s summary judgment split the issue cleanly. He ruled that using purchased books was lawful. Fair use. Transformative. No copyright infringement.
But pilfered copies? That crossed the line.
“If Anthropic had just purchased those books outright instead of using torrented files,” Ambartsumian explained, “converted those books to digital format, and then used those copies for model training, this would have been fine under Judge Alsup’s reasoning.”
Why $3,000 Could Reset AI Licensing Fees
To be crystal clear, this case won’t kill foundation AI models. It won’t force AI retraining. But it does send a pricing signal that $3,000 per work might be the new cost of doing business if you train on ill-gained books and works.
It’s a number future plaintiffs will remember.
“The biggest takeaway is that this settlement may have a larger impact on licensing fees than settlements going forward,” Ambartsumian said.
That’s the quiet part. There’s no admission of guilt. No lasting injunction. But there is a market signal. And it came with a billion-dollar receipt.
Under this agreement, Anthropic lives to train another day. Claude keeps its AI know-how intact. Only the rogue books — not the AI models — head to the shredder.