- Anthropic has been ordered by a U.S. judge to respond to claims that its expert cited a non-existent academic article in a $75 million copyright case.
- The expert’s declaration was submitted in a lawsuit alleging Anthropic used copyrighted lyrics to train its Claude AI model.
- The incident highlights growing legal scrutiny on companies like Anthropic, OpenAI, and Meta regarding the use of copyrighted material in AI training.
A federal judge in San Jose has directed Anthropic to formally address an allegation that its data scientist, Olivia Chen, referenced a fabricated academic article in court documents related to a $75 million copyright lawsuit, Reuters and Decrypt reported. The disputed citation was meant to support the company’s defense against accusations that it used copyrighted song lyrics to train its AI model Claude.
The lawsuit was brought by major music publishers including Universal Music Group, Concord, and ABKCO last year. The plaintiffs claim Anthropic unlawfully included lyrics from artists such as Beyoncé, The Rolling Stones, and The Beach Boys in Claude’s training data. According to the plaintiffs, the referenced article in Chen’s declaration was a “complete fabrication” and likely generated using Claude itself.
During a San Jose court hearing, plaintiffs’ attorney Matt Oppenheim told the judge that while he did not believe Chen fabricated the citation intentionally, “she likely used Claude itself to generate the source.” Anthropic‘s attorney, Sy Damle, countered that the issue was more likely a mis-citation rather than deliberate fabrication, and called out the plaintiffs for raising this concern late in the legal process.
U.S. Magistrate Judge Susan van Keulen described the matter as “a very serious and grave” concern and distinguished between a missed citation and “a hallucination generated by AI.” She did not grant the request to immediately question Chen, but instructed Anthropic to provide a formal response to the accusation. The company did not respond immediately to requests for comment from Decrypt.
The publishers filed the case in October 2023, seeking damages and demanding the destruction of any infringing AI training data. In its defense, Anthropic asserted in a January 2024 filing that its systems were not intended to reproduce copyrighted lyrics, describing such incidents as a “rare bug.” The publishers alleged, however, that the outputs were more frequent than claimed and that Anthropic had not provided evidence to the contrary.
This lawsuit is one among several facing AI developers. In August 2024, Anthropic was also sued by authors who said its models were trained on pirated copies of their books. Similarly, OpenAI faces lawsuits from Sarah Silverman, The Authors Guild, and The New York Times, while Meta has been named in legal actions relating to its LLaMA models allegedly trained on unlicensed works. OpenAI and Google have advocated for looser copyright rules around AI training to encourage innovation, according to their submissions for the planned U.S. “AI Action Plan.”
Meanwhile, the UK is weighing legislation that would require AI firms to disclose their use of copyrighted material after the House of Lords approved an amendment this week. These developments point to growing legal oversight of how generative AI companies like Anthropic obtain and use copyrighted data.
✅ Follow BITNEWSBOT on Telegram, Facebook, LinkedIn, X.com, and Google News for instant updates.
Previous Articles:
- AFME Urges EU Action on DLT Settlement, Collateral Eligibility
- Bitcoin Soars Near Record as Zuckerberg and Trump Fuel Crypto Surge
- Strategy Could Become Top Public Equity with $59B Bitcoin Holdings
- France to Meet Crypto Firms After Spate of Kidnapping Attempts
- Standard Chartered, FalconX Partner to Expand Institutional Crypto
