Anthropic Accused of Citing AI ‘Hallucination’ in Song Lyrics Lawsuit

An AI professional at Amazon-backed agency Anthropic has been accused of citing a fabricated educational article in a court docket submitting meant to defend the corporate towards claims that it skilled its AI mannequin on copyrighted tune lyrics with out permission.

The submitting, submitted by Anthropic information scientist Olivia Chen, was a part of the corporate’s authorized response to a $75 million lawsuit filed by Common Music Group, Harmony, ABKCO, and different main publishers.

The publishers alleged within the 2023 lawsuit that Anthropic unlawfully used lyrics from tons of of songs, together with these by Beyoncé, The Rolling Stones, and The Seaside Boys, to coach its Claude language mannequin.

Chen’s declaration included a quotation to an article from The American Statistician, meant to assist Anthropic’s argument that Claude solely reproduces copyrighted lyrics underneath uncommon and particular situations, in keeping with a Reuters report.

Throughout a listening to Tuesday in San Jose, the plaintiffs’ lawyer Matt Oppenheim referred to as the quotation a “full fabrication,” however mentioned he didn’t imagine Chen deliberately made it up, solely that she possible used Claude itself to generate the supply.

Anthropic’s lawyer, Sy Damle, informed the court docket Chen’s error seemed to be a mis-citation, not a fabrication, whereas criticizing the plaintiffs for elevating the difficulty late within the proceedings.

Per Reuters, U.S. Justice of the Peace Decide Susan van Keulen mentioned the difficulty posed “a really severe and grave” concern, noting that “there’s a world of distinction between a missed quotation and a hallucination generated by AI.”

She declined a request to instantly query Chen, however ordered Anthropic to formally reply to the allegation by Thursday.

Anthropic didn’t instantly reply to Decrypt’s request for remark.

Anthropic in court docket

The lawsuit towards Anthropic was filed in October 2023, with the plaintiffs accusing Anthropic’s Claude mannequin of being skilled on a large quantity of copyrighted lyrics and reproducing them on demand.

They demanded damages, disclosure of the coaching set, and the destruction of infringing content material.

Anthropic responded in January 2024, denying that its methods had been designed to output copyrighted lyrics.

It referred to as any such replica a “uncommon bug” and accused the publishers of providing no proof that typical customers encountered infringing content material.

In August 2024, the corporate was hit with another lawsuit, this time from authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who accused Anthropic of coaching Claude on pirated variations of their books.

GenAI and copyright

The case is a part of a rising backlash towards generative AI corporations accused of feeding copyrighted materials into coaching datasets with out consent.

OpenAI is going through a number of lawsuits from comic Sarah Silverman, the Authors Guild, and The New York Times, accusing the corporate of utilizing copyrighted books and articles to coach its GPT fashions with out permission or licenses.

Meta is called in similar suits, with plaintiffs alleging that its LLaMA fashions had been skilled on unlicensed literary works sourced from pirated datasets.

In the meantime, in March, OpenAI and Google urged the Trump administration to ease copyright restrictions round AI coaching, calling them a barrier to innovation of their formal proposals for the upcoming U.S. “AI Motion Plan.”

Within the UK, a authorities invoice that will allow synthetic intelligence companies to make use of copyright-protected work with out permission hit a roadblock this week, after the Home of Lords backed an amendment requiring AI companies to disclose what copyrighted materials they’ve used of their fashions.