A federal judge’s ruling last year in a major copyright case against Anthropic has left the legal landscape for AI training in a state of flux, with experts saying the decision signals that using copyrighted works to train models can be lawful, but the method of obtaining those works matters. Judge William Alsup ordered Anthropic to pay $1.5 billion to a group of writers whose books were used in training its AI models, but he did not rule that the training itself was illegal. Instead, Alsup found that Anthropic violated the law by acquiring those books from pirated online shadow libraries, comparing the ingestion of text by a large language model to a human writer studying literature. Attorney Cathy Gellis, who specializes in intellectual property and technology, told TechCrunch that the ruling is generally favorable for AI companies, since the judge treated the training as analogous to reading a work rather than copying it, and noted that copyright law hinges on copying, not on consumption.
The case highlights a broader legal uncertainty because copyright law has not been updated since 1976, leaving judges to interpret outdated guidelines for questions that could define the future of the AI industry. Jason Henderson, a senior attorney and founder of an IP and media practice, told TechCrunch that the law has not caught up to the reality that AI models have been trained on vast amounts of data, and courts are split in their reasoning. The central issue often comes down to fair use, which permits use of copyrighted material without permission for purposes like criticism, parody, or education, provided judges weigh factors such as the purpose of the use and its effect on the market. Henderson explained that courts tend to reject fair use when a company trains on someone’s property to directly compete with them, but they often allow it when the use does not create that kind of competition.
A separate case involving Thomson Reuters and the research firm Ross Intelligence illustrates that principle, as Judge Stephanos Bibas ruled last year that Ross’s use of Thomson Reuters’s content to build a competing AI-based legal platform was not transformative. Bibas concluded that Ross’s use did not have a further purpose or different character from the original content’s purpose, making it an unfair use. While authors might argue that chatbots compete with them by generating synthetic books, that argument has not yet succeeded in court, leaving the door open for future litigation.
Gellis also distinguished between copyright issues in AI training and copyright protections for AI-generated output, pointing to the case of Thaler v. Perlmutter, where a court ruled that a wholly AI-generated work cannot be copyrighted. That decision raises practical questions about how to prove whether a work was generated by AI and what percentage of it involved human assistance. Gellis compared the situation to writing a novel in Microsoft Word and running spell check, noting that people feel comfortable saying the software does not own the novel, but AI is forcing a reexamination of assumptions that have gone unchallenged.
Most AI companies remain entangled in ongoing litigation over these issues, meaning a definitive resolution is unlikely in the near term. Gellis told TechCrunch that early rulings are influencing the direction of the industry, but those decisions could be overturned if other courts reach different conclusions, and later stages of litigation will determine which interpretation prevails. She added that it would be unwise for AI companies to ignore these rulings while they continue to shape the legal and commercial environment.
More AI news from TechManNews.
