Anthropic Hit with Copyright Lawsuit Over LLM Training

Three writers on Monday filed a class-action lawsuit against artificial intelligence firm Anthropic, saying the company misused their works and many more writers’ material to train its generative artificial intelligence chatbot, Claude.

Writers and journalists Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson say Anthropic used their work to teach its large language model, Claude, to respond to human prompts.

“Anthropic styles itself as a public benefit company, designed to improve humanity,” the lawsuit reads. “For holders of copyrighted works, however, Anthropic already has wrought mass destruction … 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 is now one of several aimed at protecting copyright holders, including writers, visual artists, news outlets, and recording artists over the training of models on copyrighted works. Different groups have sued OpenAI and Meta. The New York Times was one of they highest profile plaintiffs looking to sue OpenAI, saying the company used million of its articles without permission in model training.

OpenAI has claimed that the training was protected by fair use copyright law. But it since forged dozens of new media partnerships to train the large language model behind its blockbuster ChatGPT chatbot. The Atlantic, BuzzFeed, News Corp, the Associated Press, Vox Media, and Reddit are some notable partners that gave OpenAI legal access to copyrighted articles, images, and more.

Related:More Newspapers Join the Copyright Battle Against OpenAI, Microsoft

This is the second case against Anthropic. Music publishers filed a lawsuit last year against the Amazon-backed AI firm. In the case filed Monday, the authors claim Anthropic “built a multi-billion-dollar business by stealing hundreds of thousands of copyrighted works.”

The lawsuit requests monetary damages and an order to permanently block Anthropic from misusing the authors’ work in the future.

InformationWeek has reached out to Anthropic for comment and will update.