A legal team representing a former Homeland Security official apologized after fabricating quotes attributed to Anthropic's Claude AI in a court filing related to Trump administration layoffs. The attorneys used Claude Console to assist in drafting the document, but the AI-generated quotes inserted into the filing did not match any actual statements Claude produced.
The error surfaced when the opposing counsel challenged the authenticity of the quoted material. The lawyers acknowledged they misrepresented Claude's responses, claiming the quotes were meant to illustrate hypothetical concerns rather than direct output. The incident highlights the risks of deploying large language models in legal contexts without rigorous verification protocols.
Anthropic, the company behind Claude, did not immediately respond to requests for comment on how its AI was misused. The case involves terminations stemming from executive orders issued during the Trump administration targeting federal employees deemed part of a "deep state."
This episode marks the second major instance of hallucinated AI quotes appearing in court filings. Earlier cases saw attorneys rely on ChatGPT-generated content that contained false citations and non-existent case law. Legal ethics boards across multiple jurisdictions have begun issuing guidance warning lawyers that AI tools cannot be trusted to generate accurate citations or quotes without independent verification.
The incident underscores a growing tension between adoption of AI tools for efficiency and the requirement for factual accuracy in legal documents. Courts and bar associations increasingly expect attorneys to independently verify all material generated by language models before submission. Failure to do so exposes lawyers to sanctions, malpractice liability, and potential disciplinary action.
The case remains ongoing, but the credibility damage to the legal team could influence how judges evaluate other arguments presented in the filing. This development serves as a cautionary tale for legal professionals integrating generative AI into their workflows without establishing strict fact-checking procedures.