Last year, a federal judge ordered Anthropic to pay $1.5 billion to a group of authors whose works were used to train its AI models — but the ruling wasn’t the clear victory for writers that headlines suggested. Judge William Alsup found that Anthropic’s AI training itself was lawful under fair use principles. The penalty came instead because the company sourced training data from illegal online shadow libraries.
“Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them — but to turn a hard corner and create something different,” Alsup wrote, drawing an analogy between machine learning and a writer’s study of literature.
Also read: OpenAI gains on Anthropic in business spending, new Ramp data shows
The decision highlights a widening gap between the rapid evolution of AI technology and a copyright framework that hasn’t been meaningfully updated since 1976. As judges grapple with how to apply half-century-old guidelines to trillion-parameter models, the legal environment remains fragmented and uncertain.
Fair use is the battleground for AI training disputes
At the center of these disputes is fair use law, which permits limited use of copyrighted material without permission for purposes like criticism, commentary, parody, and education. Courts weigh four factors when evaluating fair use claims: the purpose and character of the use, the nature of the copyrighted work, the amount used, and the effect on the market.
Also read: AI accounting startup Rillet raises $100M, hits unicorn status in 48 hours
“Copyright is always about protecting and growing the market,” Jason Henderson, Senior Attorney and Founder of the IP & Media Practice at JWL International, told TechCrunch. “The courts are kind of all over the place in their reasoning [in AI cases]. What’s tending to win is if what you’re doing is you’re training on somebody’s property because your purpose is to directly compete, then the courts will frown on it… If what you’re doing is not going to compete, then the courts are tending to find ways that it will be okay.”
That distinction was central to a separate case where Thomson Reuters sued Ross Intelligence, a legal research firm that used Reuters’ content to build a competing AI-powered platform. Judge Stephanos Bibas ruled against Ross last year, writing that its use was “not transformative because it does not have a ‘further purpose or different character’ than Thomson Reuters’s.”
The contrast between the two rulings offers a rough roadmap: training on copyrighted works to create a directly competing product is risky, while training that produces something new may be defensible. But authors have yet to successfully argue that general-purpose chatbots are direct competitors to their books.
Why the $1.5 billion fine may be a win for AI companies
Attorney Cathy Gellis, who specializes in intellectual property and technology law, believes the Anthropic ruling ultimately favors AI developers despite the hefty penalty. “I think it is generally good news for AI training that he looked at what was going on and really sort of thought it analogous to reading a copyrighted work as opposed to copying a copyrighted work,” she told TechCrunch.
“Copyright law hinges on copying, but it doesn’t hinge on using the work or experiencing the work, consuming the work, reading the work.”
The math also favors the industry. A $1.5 billion settlement, while substantial, is a fraction of Anthropic’s projected revenue trajectory — the company anticipates around $200 billion in annual revenue by 2028. For well-funded AI labs, the cost of settling may be far less than the cost of negotiating licenses for every copyrighted work in their training datasets.
The unresolved question of AI-generated content
Beyond training, courts are also wrestling with whether AI-generated works can be copyrighted at all. In Thaler v. Perlmutter, a court ruled that works created entirely by AI are not eligible for copyright protection. That decision raises thorny questions about how to determine whether a work was AI-generated, and what percentage of human involvement is sufficient to claim authorship.
“If you write your novel in [Microsoft] Word and run spell check, we kind of feel comfortable with the idea of saying that Word does not own your novel,” Gellis said. “[AI] is forcing us to look at a whole bunch of decisions that we kind of ignored for a while.”
Most major AI companies remain mired in pending litigation over these issues, meaning a definitive legal resolution is unlikely in the near term. The early rulings, however, are already shaping industry behavior and legal strategy.
“What you are seeing is that the initial opening volleys are being influential, and that influence itself could be undone if other courts decide different things, and it’ll take later states of litigation to figure out which one will prevail,” Gellis said. “But in the meantime, all these decisions are shaping everything that’s happening. It would be kind of foolish for the AI companies to ignore them.”
For authors and publishers, the path forward remains unclear. While the Anthropic ruling penalized the use of pirated materials, it did not establish that authors have a right to compensation when their works are legally obtained and used for training. Until higher courts or Congress provide clearer guidance, the legality of AI training on copyrighted books will remain a case-by-case question — and one of the most consequential legal debates in the technology industry.
This article is for informational purposes only and does not constitute legal or financial advice. The legal space surrounding AI and copyright is rapidly evolving and subject to change.
