Anthropic has won a significant legal victory regarding artificial intelligence and copyright. A federal judge has decided that Anthropic can train its AI models on legally purchased books without needing permission from authors, classifying this practice as fair use. However, the ruling is limited to physical books that the company has bought and digitized. Judge William Alsup of the Northern District of California also indicated that Anthropic faces a separate trial regarding the alleged piracy of millions of books from the internet. This landmark decision allows Anthropic to train its AI tools on copyrighted materials without permission, a ruling deemed transformative by Senior District Judge William Alsup. However, the case is not over; the court also allowed authors to take Anthropic to trial for allegedly pirating over seven million copies of books for its training database. The judge highlighted that while training on copyrighted works was lawful, the use of pirated materials was not excusable, indicating potential damages could reach billions of dollars.
Anthropic spent millions acquiring used print books, which were then destructively scanned into digital files, leading to the disposal of the original copies. Alsup noted that while the transformative nature of using purchased books for training could qualify as fair use, the act of downloading pirated versions clearly crossed a legal boundary.
Meta Platforms has secured a significant legal victory regarding the use of copyrighted materials for artificial intelligence training, with a federal judge ruling in the company’s favor. However, this ruling does not confirm that such use qualifies as fair use, as highlighted by Judge Vince Chhabria. The lawsuit was initiated by 13 authors who claimed that Meta illegally trained its AI systems on their works without permission. Judge Chhabria stated that while Meta is entitled to summary judgment on its fair use defense, the plaintiffs failed to present compelling arguments to support their claims. The judge specifically noted that the plaintiffs did not sufficiently demonstrate how Meta’s actions would dilute their licensing opportunities in the market.
Why do we care?
We care because these rulings signal that the legal terrain for training AI on copyrighted content is shifting—but it’s far from settled. For IT services firms, especially those integrating AI into client solutions, the implications are both strategic and operational. These rulings are a strategic green light—but with a flashing yellow for anyone downstream of the AI supply chain. MSPs and IT services firms must begin treating AI data provenance like software licensing: a risk to be audited, verified, and disclosed. Those who blindly adopt “black box” models will find themselves exposed—legally, ethically, and reputationally.