Anthropic sued for training its chatbot using books

2024-08-21

A group of authors is suing artificial intelligence startup Anthropic, accusing it of committing "massive theft" by training chatbots with pirated copyrighted books. Anthropic, headquartered in San Francisco and founded by former OpenAI executives, aims to create chatbots that are more trustworthy than ChatGPT and serve as a core part of Claude's marketing, promising to provide "reliable, interpretable, and controllable AI systems." The company has even called for AI proposals with a focus on safety. However, a lawsuit filed on Monday (August 19) in the San Francisco federal court has somewhat tarnished this image, alleging that Anthropic used a library of pirated works to train its AI products. Similar accusations have been made against OpenAI's ChatGPT and a music publisher associated with Anthropic. According to the Associated Press, the lawsuit states, "It is no exaggeration to say that Anthropic's model seeks to profit by plundering the human expression and creativity behind these works." Who filed the lawsuit against Anthropic? The authors filing the lawsuit include Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, representing a group of fiction and non-fiction writers with similar concerns. The case accuses the tech company of using human works to build its AI models. This helps train AI chatbots like Claude to generate text, but without the consent of the original authors or providing them any compensation. While this case focuses on written works, Anthropic also faces similar accusations from music publishers, who claim that generative AI is profiting from the misuse of creative works. In response, Anthropic and other tech companies defend themselves by arguing that AI model training falls under the fair use principle in US copyright law. This typically covers educational, research, or transformative uses of copyrighted works. However, the lawsuit against Anthropic argues that the learning process of AI systems is different from that of humans. The lawsuit states, "When humans learn by reading books, they either purchase legitimate copies or borrow from libraries that have purchased these books, thereby providing at least some compensation to authors and creators."