A lawyer representing Anthropic, the AI company behind the Claude chatbot, told a U.S. federal court that her law firm, Latham & Watkins, was responsible for an AI-generated error in a legal filing tied to a copyright lawsuit from major music publishers.
Attorney Ivana Dukanovic admitted that she used Anthropic’s own chatbot Claude to create a citation, which included a fake article title and author names.
The AI “hallucination” occurred while preparing an expert report in a lawsuit filed by Universal Music Group, Concord, and ABKCO. These publishers accuse Anthropic of using song lyrics without permission to train its AI systems.
While the publication year and link were correct, the AI-generated citation included inaccurate details that were not caught before filing. Dukanovic called it an “embarrassing and unintentional mistake” in a filing submitted Thursday.
The issue came to light when the publishers’ attorney, Matt Oppenheim, questioned whether Anthropic’s data scientist had relied on a fake source. During a court hearing, Judge Susan van Keulen called it a “serious and grave issue,” noting the difference between a missed citation and one fabricated by AI.
Anthropic’s legal team later clarified that the underlying source was real, but the hallucinated details were added by Claude and not caught in time. Latham & Watkins says it has since added “multiple levels of review” to avoid such mistakes.
This case is one of many lawsuits testing how companies use copyrighted content to train AI models, a legal gray area attracting growing scrutiny.
Leave a comment