Judge rebukes DOJ for fabricated legal citation in immigration case
The U.S. Department of Justice filed a legal brief containing a fabricated case citation that a federal judge found was likely generated by artificial intelligence. The judge warned the DOJ to ensure future filings do not include non-existent case law but declined to impose sanctions.
What happened
In May 2026, the U.S. Department of Justice filed a legal brief in an immigration bond case involving an ICE detainee named Izzeddin Daghra. The brief cited a Sixth Circuit case called Taylor v. Hott to argue the court lacked authority to intervene in the government's appeal of a $35,000 bond decision.
Chief U.S. District Judge Hala Jarbou of the Western District of Michigan found the citation did not exist. The cited page of the Federal Appendix actually contained an unrelated ruling about commercial arbitration, and no federal case with the caption Taylor v. Hott could be found anywhere. The judge determined the citation 'seems likely produced by generative artificial intelligence.'
Reported impact
The judge issued a warning that the government must ensure future filings do not include non-existent case law but did not impose sanctions at that time.
Index analysis
Generative AI models can produce fabricated citations that appear plausible but do not correspond to actual legal cases, a phenomenon known as 'hallucination.' The cited case Taylor v. Hott did not exist in any federal reporter.
Governance controls
The judge noted that 'any attorney who uses AI must scrupulously review its work product to ensure that the cited cases exist and that the citations accurately and fairly represent the underlying case law,' emphasizing the duty of candor required of government counsel.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.