Copyright "Theft" is Mostly Legal
Anthropic can't be forced to pay for 93 percent of the books it pirated
A number of people, including us here at AI Central, had some obvious questions about the recently announced settlement that only covers 500,000 of the 7 million books that Anthropic admitted pirating to train its excellent generative-AI systems.
The reason is very simple: Anthropic doesn’t have to pay anything for 93 percent of the books it pirated because the US courts simply won’t do anything about the infringement of a copyright for any intellectual property that isn’t formally registered with the U.S. Copyright Office.
In addition to establishing a public record of a copyright claim, registration offers several other statutory advantages:
• Before an infringement suit may be filed in court, registration is necessary for U.S. works.
• Registration establishes prima facie evidence of the validity of the copyright and facts stated in the certificate when registration is made before or within five years of publication.
• When registration is made prior to infringement or within three months after publication of a work, a copyright owner is eligible for statutory damages, attorneys’ fees, and costs.
In fact, the US legal system won’t even protect any work that is pirated after the copyright registration is filed, but before the copyright registration is approved by the copyright office. This is well-established law, not merely black letter law, but case law laid down by the very highest legal authority in the land, the U.S. Supreme Court.
Pending Application is Not Enough - Copyright Registration is a Prerequisite to Filing Suit
March 6, 2019
The U.S. Supreme Court rendered a unanimous decision in an opinion penned by Justice Ruth Bader Ginsburg regarding whether 17 U.S.C. § 411’s requirement that “registration” must be “made” requires the Copyright Office to issue a registration (the “registration approach”), or merely requires plaintiff to apply for one (the “application approach”) before a lawsuit is filed.
Fourth Estate Public Benefit Corp. v. Wall-Street.com, LLC , No. 17-571 (S. Ct. March 4, 2019).
In other words, there is no way for the author of any work that is published and copyrighted outside the United States, or is published but not registered with the US copyright office, to pursue legal redress against Anthropic, Meta, or any other AI company even when the company admits to having pirated the author’s works.
Now, it’s certainly possible that Anthropic intends to compensate the authors of the 9 out of 10 pirated works that are not included in the settlement; perhaps it’s not an accident that the ratio of money set aside for the settlement to new investment (8.66) is not dissimilar to the ratio of registered works to unregistered works (14). But that’s by no means a given and could merely be a coincidence.
There are still various means of legal redress that can be pursued by certain rights-holders. Foreign publishers can certainly take Anthropic and the other AI companies to court in their home countries with an expectation of success, although that’s only likely to prove financially worthwhile if the AI company sued happens to have a physical presence there. But the vast majority of self-publishers and foreign authors are going to be left out of luck, at least with regards to this initial settlement.
The upshot is that large-scale book piracy by the AI companies appears to remain effectively legal and financially viable in the United States, and despite its historic size, the Anthropic settlement is not going to do anything at all for the overwhelming majority of published authors whose works were pirated.
If you want to find out if you have any registered copyrights that will permit your pirated work to be included in the Anthropic settlement, you can search the records of the US Copyright Office here by entering your name or the title of your book in the search box.
Elon Musk's Department of Government Efficiency (DOGE) has been linked to significant actions affecting the U.S. Copyright Office in May 2025. President Donald Trump fired the head of the U.S. Copyright Office, Shira Perlmutter, shortly after the office released a report stating that training artificial intelligence systems on copyrighted works may not qualify as fair use under U.S. copyright law.
Democratic Representative Joe Morelle suggested the timing was not coincidental, implying the firing was connected to Perlmutter's refusal to allow Elon Musk to use copyrighted material for AI training without compensation.
Although reports of a physical standoff between Capitol Police and DOGE at the Copyright Office circulated, these appear to be unconfirmed or metaphorical, with one Reddit user noting "doesn’t look like there is a physical standoff with DOGE".
However, DOGE's broader influence is evident in a series of leadership changes and firings at federal institutions, including the Library of Congress and the U.S. Copyright Office, which music organizations and lawmakers have criticized as an overreach by the Executive Branch into areas constitutionally under Legislative Branch authority.
The firings have raised concerns among creators and legal experts that DOGE's actions could weaken copyright protections, potentially benefiting large tech companies seeking to use copyrighted content for AI development without paying licensing fees.
Critics, including members of Trump’s own circle, believe Musk and White House AI and Crypto Czar David Sacks were behind the moves to reshape intellectual property policy in favor of tech interests.
However, the acting officials appointed—Paul Perkins, Brian Nieves, and Todd Blanche—are described as staunch critics of Big Tech, suggesting the administration's motives may be more complex.