In October 2023 the music publishing companies Universal Music Group, Concord, and ABKCO filed a federal lawsuit against the artificial intelligence company Anthropic, accusing them of violating copyright law by using the lyrics to over 500 songs to train their chatbot. Now, a federal judge is ordering Anthropic to respond to allegations that they submitted a court filing containing a "hallucination" created by AI as part of its defense against those copyright claims, Reuters reports.
During a hearing Tuesday in California, a lawyer representing those three publishers accused an Anthropic data scientist of citing a nonexistent academic article in an effort to corroborate their argument that their chatbot, called Claude, does not violate copyright law by responding to human prompts with existing lyrics and chord progressions from the publishers' music. (For example, the suit argues that if you ask Claude to write a song about the death of Buddy Holly, it might respond with lyrics to Don McLean's "American Pie" without proper disclosure.)
The forged article in question appeared to be from the journal American Statistician, and it was used to argue that it's a "rare event" for Claude to regurgitate existing copyrighted material. The publishers' attorney said during the hearing that one of the supposed authors told him the article was a "complete fabrication."
Anthropic was founded in 2021 and has since received financial backing from Amazon, Google, and the crypto entrepreneur Sam Bankman-Fried, who was convicted of fraud in November 2023.
UPDATE: On Thursday (May 15) Anthropic admitted that Claude produced an erroneous citation that its lawyers overlooked, but called it "an honest citation mistake and not a fabrication of authority."
