Lawyers for a former Homeland Security official have acknowledged they used Anthropic’s Claude Console to help draft a court filing — and that the AI-generated text included fabricated quotes. The admission came in a case stemming from layoffs during the Trump administration, raising questions about the use of generative AI in legal documents.
How the AI was used
Attorneys representing the ex-official told the court they relied on Claude Console, a tool from the AI startup Anthropic, to prepare a motion. They said they did not catch that the software inserted quotes that were not from any actual source. The filing was part of litigation over personnel actions taken during the Trump-era workforce reductions.
The admission was made in a filing with the court, where the lawyers explained that they used the AI tool for efficiency but failed to verify the accuracy of the citations. They did not name which specific attorney or team member prompted the AI, nor did they specify how much of the document was generated by Claude.
The fabricated quotes
The disputed quotes appeared in a legal brief submitted to the court. The lawyers later realized the passages attributed to individuals were not real — they were invented by the AI model. The fabricated language was removed after the error was discovered, but the incident has drawn scrutiny from the opposing counsel and the presiding judge.
Court records show the case involves a former Homeland Security official who challenged his termination during the Trump administration’s broader federal workforce cuts. The lawyers’ admission that an AI tool produced false content in a sworn filing has become a separate point of contention.
Broader implications for AI in legal work
This is not the first time a legal team has been caught using generative AI to produce citations or quotes that turned out to be fake. Earlier cases involved lawyers who used ChatGPT to draft motions containing nonexistent case law. The Anthropic incident underscores that the problem extends beyond one AI provider.
Judges and bar associations have started issuing guidance on when and how attorneys may use AI tools. Some courts now require lawyers to certify that no AI-generated content was submitted without human verification. The admission in this case could lead to sanctions or a referral to disciplinary authorities, though no such action has been announced yet.
What happens next
The court has not yet ruled on the fallout from the fabricated quotes. The opposing side may ask for the filing to be stricken or for the lawyers to be held in contempt. The former Homeland Security official’s legal team has said it will take steps to ensure no further AI-generated errors occur. A hearing on the matter is expected in the coming weeks.
