The erroneous citation was included in an expert report by Anthropic data scientist Olivia Chen last month defending claims about the company using copyrighted lyrics to train Claude, Anthropic’s large language model. Anthropic is being sued for alleged misuse of copyrighted materials to train its generative AI tools.
Although the citation carried the correct link, volume, page numbers, and publication year, the LLM, known as Claude, provided a false author and title, according to a declaration from Ivana Dukanovic, an associate at Latham & Watkins LLP and attorney of record for Anthropic.
In the declaration, Anthropic attorney Dukanovictook accountability for the mishap, saying it was “an honest citation mistake and not a fabrication of authority,” according to the filing.
She said the Latham & Watkins team found the article as “additional support for Ms. Chen’s testimony.” Then, Dukanovic asked Claude “to provide a properly formatted legal citation” for the article, which resulted in the hallucinated sourcing..
Claude did not complete the citation correctly, and the attorney’s “manual citation check did not catch that error,” according to Dukanovic.
“This was an embarrassing and unintentional mistake,” Dukanovic said.
