Fact Checked by 
A group of authors is suing AI startup Anthropic, claiming it committed ‘large-scale theft’ by training its chatbot on pirated copies of copyrighted books.
Based in San Francisco and founded by ex-OpenAI bosses, Anthropic made being the more trustworthy version of ChatGPT a central part of its marketing efforts around its chatbot, Claude, promising to deliver “reliable, interpretable, and steerable AI systems.” The company even made a point of calling for safety-focused AI proposals.
That image has been somewhat shattered by the lawsuit filed on Monday, August 19 in a San Francisco federal court, alleging that Anthropic has taught its AI product using libraries of pirated works. Similar claims have been made against OpenAI’s ChatGPT and against Anthropic itself by music publishers.
“It is no exaggeration to say that Anthropic’s model seeks to profit from strip-mining the human expression and ingenuity behind each one of those works,” the lawsuit states, as reported by AP News.
Anthropic has not yet made a public statement in response to the lawsuit at the time of writing.
What lawsuit has been brought against Anthropic?
The writers leveling the lawsuit are Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, seeking to represent a class of similarly-minded fiction and non-fiction authors. The case alleges that the tech company has built its AI models using human work.
This helps to train AI chatbots like Claude to produce text, without the consent of the original creators and without offering them any compensation. While this case focuses on the written word, Anthropic is also currently facing similar claims from music publishers claiming that generative AI is profiting from the misuse of creative works.
In response, Anthropic and other tech companies have defended themselves by arguing that AI model training falls under the fair use doctrine of US copyright law. This typically covers teaching, research, or transforming copyrighted works into something new.
However, the lawsuit against Anthropic argues that AI systems do not learn the way humans do, writing: “Humans who learn from books buy lawful copies of them, or borrow them from libraries that buy them, providing at least some measure of compensation to authors and creators.”
Featured image: Anthropic
