The copyright case for human AI includes “guardrails” to the song’s lyrics

A federal judge has approved a temporary agreement in a lawsuit alleging that artificial intelligence company Anthropic used copyrighted song lyrics without permission or payment while training its system.

Music publishers, including Universal Music Group, Concord Music Group and ABKCO, have filed a lawsuit Copyright infringement claim v. Anthropic in 2023. The lawsuit alleges that Anthropic Claude I The chatbot was trained on the lyrics of at least 500 songs by major artists, including Beyoncé and Katy Perry, and produced results for queries about them, without obtaining permission or compensating the rights holders.

U.S. District Judge Eomi Lee signed off on Thursday deal between the publishers and Anthropic that requires the AI ​​company to maintain existing copyright guardrails — which means not submitting old or new song lyrics — throughout the litigation process.

The agreement states that nothing prevents Anthropic from “expanding, improving, enhancing or changing the implementation of these guardrails, provided that such changes do not materially reduce the effectiveness of the guardrails.” It also allows publishers to notify Anthropic at any time if they believe the guardrails are inadequate, at which point the company must address their concerns.

The court is expected to decide in the coming months whether songs owned by music publishers can be used to train artificial intelligence systems.

The case highlights the tension between AI developers and copyright holders, and not just in the music industry. Publishers, including the New York Times, have sued OpenAI and Microsoft, the maker of ChatGPT, and the Wall Street Journal and Dow Jones have also sued Perplexity, all of which Claim of copyright infringement. Meanwhile, other publishers have struck licensing deals with AI companies including OpenAI and Meta.

Anthropic, which is backed by Amazon, Google, Salesforce and others, was founded by former OpenAI researchers in 2021. Its system is trained on datasets that may include copyrighted material such as song lyrics, raising questions about fair use and proper licensing.

In a statement shared with CNET, the company affirmed its commitment to copyright compliance, including by entering into the current agreement.

“Cloud is not designed to be used in copyright infringement, and we have several processes in place designed to prevent such infringement,” a company spokesperson said. “We continue to look to demonstrate that, consistent with current copyright law, the use of potentially copyrighted materials in training generative AI models is substantive fair use.”

Universal Music Group, Concord Music Group and ABKCO did not immediately respond to requests for comment, but the original lawsuit alleges that Anthropic is “illegally copying and disseminating vast amounts of copyrighted works.”

“Just like developers of other technologies that came before, from the printing press to the mimeograph to the web crawler, AI companies must follow the law,” the lawsuit said.

Anthropic still faces other legal challenges. Earlier this year, the company was named in a copyright infringement class action lawsuit from authors alleging that Anthropic “built a multibillion-dollar business by stealing hundreds of thousands of copyrighted books.”