Anthropic says Claude AI caused “embarrasing and unintentional” error in lawsuit

Anthropic on Thursday admitted that a faulty reference in a court paper was the result of its own AI assistant Claude and called the lapse “an embarrassing and unintentional mistake.”

Universal, Concord, and ABKCO sued Anthropic last year, alleging that the San Francisco-based company copied their lyrics wholesale to train Claude. The case is one of several high-profile cases over whether technology firms may use copyrighted work without permission to train AI systems.

Reuters reported that during a hearing on Tuesday, attorney Matt Oppenheim, who represents Universal Music Group, ABKCO, and Concord, told the court that the chatbot’s quotation was a “complete fabrication” to support Anthropic’s case.

U.S. Magistrate Judge Susan van Keulen called the latest allegation “a very serious and grave issue” and stressed “a world of difference between a missed citation and a hallucination generated by AI.” The judge ordered the parties to explain the matter fully in their written submissions.

Anthropic lawyers say Claude AI messed up while formatting

In a written response on Thursday, Anthropic lawyer Ivana Dukanovic said the reference was real but was mishandled when Claude was asked to format it in legal style. She explained that wrong volume and page numbers produced by the bot were fixed during a manual review, yet wording errors slipped through without notice.

“Although providing the correct publication title, publication year, and link to the provided source, the returned citation included an inaccurate title and incorrect authors,” Dukanovic wrote, adding that the glitch was not a “fabrication of authority.”

Anthropic apologized for the confusion and said the slip was neither deliberate nor an attempt to mislead the court.

AI mistakes keep causing trouble in court

Several lawyers have recently faced criticism or sanctions for relying on AI-generated material that turned out to be false. In her filing, Dukanovic said the law firm Latham & Watkins, which represents Anthropic, has added “multiple levels of additional review to work to ensure that this does not occur again.”

The incident joins a growing list of courtroom errors linked to generative AI.

Last week, a California judge scolded two law firms for failing to say they had used AI to craft a supplemental brief packed with “bogus” references that “didn’t exist.” In December, a misinformation researcher admitted that ChatGPT had invented citations in a filing he prepared.