Anthropic accused of quoting AI ‘hallucination’ in lyrics lawsuit

An AI expert at the company-supported company Anthropic has been accused of quoting a manufactured academic article in a court application that is intended to defend the company against claims that it has trained its AI model on copyright protected lyrics without permission.

The submission, submitted by anthropic data scientist Olivia Chen, was part of the company’s legal response to a $ 75 million lawsuit that was filed by Universal Music Group, Concord, Abkco and other large publishers.

The publishers claimed in the 2023 trial that anthropic illegally used texts of hundreds of songs, including those of Beyoncé, The Rolling Stones and The Beach Boys, to train his Claude language model.

Chen’s statement included a quote for an article of The American statisticianintended to support Anthropic’s argument that Claude only reproduces copyright texts under rare and specific circumstances, according to a Reuters report.

During a hearing on Tuesday in San Jose, the lawyer of the claimants Matt Oppenheim called the quote a ‘full manufacture’, but said he did not believe that Chen deliberately came up with, only that she probably used Claude herself to generate the source.

Anthropic’s lawyer, Sy Damle, told Chen’s fault, seemed to be a miscitation, not manufacturing, while he criticized the claimants to call on the issue in the procedure.

Per Reuters, The American magistrate judge Susan van Keulen said that the issue was “a very serious and serious” care, and noted that “there is a world of difference between a missed quote and a hallucination generated by AI.”

She refused a request to immediately question Chen, but ordered anthropic to formally respond to the allegation on Thursday.

Anthropic did not respond immediately Decrypts Request for comments.

Anthropic in court

The lawsuit against anthropic was brought in October 2023, with the claimants accusing the Claudi Model of Anthropic to be trained at a huge volume of copyrighted texts and to reproduce them on request.

They demanded damage, disclosure of the training set and the destruction of infringing content.

Anthropic responded in January 2024 and denied that his systems were designed to perform copyright texts.

It called such a reproduction a “rare bug” and accused its publishers not to provide evidence that typical users encountered infringing content.

In August 2024 the company was hit with another lawsuit, this time from authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson, who accused Anthropic of training Claude on Pirated versions of their books.

Genai and copyright

The case is part of a growing return against generative AI companies that are accused of feeding copyrighted material in training datas sets without permission.

OpenAi is confronted with several lawsuits of comedian Sarah Silverman, the Authors’ guild, and The New York TimesAccusing the company of the use of copyright books and articles to train its GPT models without permission or licenses.

Meta is mentioned in similar suits, with claimants who claim that his LLA models are trained on literary works without a permit from illegal data sets.

In the meantime, OpenAi and Google insisted on the Trump administration in March to alleviate copyright restrictions on AI training and to call them an obstacle to innovation in their formal proposals for the coming American ‘AI Action Plan’.

In the United Kingdom, a bill became that artificial intelligence companies would enable it to use a roadblock protected with copyrights this week, after the House of Lords supported an amendment that AI companies forced to reveal what copyright material they used in their models.