AI Developer Filed a Hallucinated Citation in Its Own Copyright Defense
What happened
The company at the center of a high-profile copyright lawsuit brought by Universal Music Group, Concord, and ABKCO found itself facing a second, self-inflicted problem in May 2025. The underlying suit, filed in San Jose, California, accused Anthropic of using copyrighted song lyrics to train its Claude AI model without authorization. While defending against those claims, Anthropic's legal team submitted a court filing that cited an academic article from The American Statistician journal. The article did not exist.
The citation appeared in a declaration submitted by Olivia Chen, a data scientist working for Anthropic. It listed a title and named authors that were entirely fabricated. Nothing in the filing distinguished the citation from a legitimate reference, and it passed through the company's legal team into the court record without being checked against the journal itself.
The error surfaced when attorneys for the plaintiffs contacted the supposed authors and the journal directly. Both confirmed the article was fictitious. The judge, on learning of the fabrication, described the matter as "very serious and grave" and required Anthropic to formally respond to the court.
Anthropic's lawyers took responsibility and attributed the mistake to a workflow failure: the attorney preparing the filing had used the company's own AI chatbot to help format references, and the tool had hallucinated the title, authors, and publication details. The legal team described the result as an "embarrassing and unintentional mistake," framing it as a citation error that slipped past manual review rather than a deliberate fabrication.
The incident exposes a gap that is easy to miss in fast-moving legal work. An AI tool and a human reviewer occupied the same production step, but neither was required to produce evidence that the citation had been independently verified against a real source. Once the filing was submitted, there was no trail showing who checked what, or when. A provable record of what a system generated, with a documented verification step tied to it, would have made the failure visible before it reached a judge rather than weeks later when opposing counsel did the check that should have happened first.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.