Anthropic, a generative artificial intelligence (GenAI) company, has argued in a US court that using copyrighted content in the training data of large language models (LLMs) should be considered “fair use.” The company claims that without this use of copyrighted material, general-purpose AI tools like theirs would not be possible. However, music publishers, including Concord, Universal Music Group, and ABKCO, have initiated legal action against Anthropic, alleging wide-scale infringement of their copyrighted song lyrics. They argue that reproducing and distributing copyrighted works without permission is unlawful, regardless of whether it is done by an AI company. The publishers are seeking damages, accounting of training data and methods, and the destruction of infringing copies. In response, Anthropic contends that the use of copyrighted material in training its AI model Claude is lawful, as it is unrelated to the expressive purpose of the original works. Anthropic believes that requiring licences for the use of copyrighted content would impede the development of generative AI models and disproportionately benefit only the wealthiest entities. They also argue that amassing a training corpus through arm’s-length licensing transactions is unfeasible. The company asserts that their dataset consists of publicly available internet data, including song lyrics, and any inclusion of copyrighted material is unintentional. While Anthropic claims to have implemented safeguards to prevent reproduction of copyrighted material, they admit that their measures may not be perfect, particularly if a user intentionally misuses their AI model. Similar copyright cases have been filed against other AI companies such as OpenAI, Stability AI, Microsoft, Google, and Meta. No court decisions have been made, but the outcomes could establish precedents for future technology development. The American Society of Composers, Authors, and Publishers (ASCAP) also argues that unauthorized and non-personal use of copyrighted material for training generative AI models does not qualify as fair use and requires consent from copyright holders.
Menu
