Attorneys used AI tools to generate legal research for a court brief, resulting in fabricated case citations and quotations that misled a judge and led to $31,100 in sanctions.
In the case of Lacey v. State Farm in California, attorneys from two law firms (Ellis George and K&L Gates) submitted a supplemental brief to Special Master Judge Michael Wilner that contained AI-generated legal research with numerous false citations and quotations. Attorney Mr. Copeland at Ellis George used various AI tools to generate an 'outline' for the brief containing problematic legal research, which was then incorporated into the final brief by K&L Gates without verification. Of the 27 legal citations in the ten-page brief, approximately nine were incorrect in some way, with at least two cited authorities that do not exist at all. Several quotations attributed to judicial opinions were fabricated and did not accurately represent the materials. When Judge Wilner initially flagged two incorrect citations, the attorneys resubmitted a revised brief that still contained half a dozen AI-generated errors without disclosing the use of AI tools. The judge concluded that the attorneys acted in bad faith through their undisclosed use of AI, failure to cite-check the research, and resubmission of defective materials without proper disclosure. As sanctions, the judge struck the briefs, denied discovery relief, and ordered the law firms to pay $31,100 in compensation to the defense and Special Master fees.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that inadvertently generate or spread incorrect or deceptive information, which can lead to inaccurate beliefs in users and undermine their autonomy. Humans that make decisions based on false beliefs can experience physical, emotional or material harms
Human
Due to a decision or action made by humans
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed