Hallucinations & Sanctions

The British Columbia Supreme Court addressed AI hallucinations by self-represented plaintiffs in costs submissions following a successful civil judgment. The plaintiffs cited non-existent cases in their costs arguments, leading Justice Blok to address the AI-generated fabrications in a published costs judgment — one of the first Canadian provincial court decisions to do so.

A Northern District of California court sanctioned a pro se plaintiff after she cited fictitious cases — and then submitted a false explanation to the court, claiming the fabricated citations resulted from ‘editing errors.’ The court found this explanation ‘simply not credible’ and that the plaintiff exacerbated her Rule 11 breach by certifying the fictitious cases as real.

A Pennsylvania federal court addressed large language model (LLM) hallucinations in an immigration detainee’s habeas petition while ordering a bond hearing for the Ghanaian permanent resident. The case is notable for its use of ‘LLM’ terminology in a published immigration decision.

The Arizona Court of Appeals affirmed dismissal of a pre-litigation discovery petition while refusing to consider case citations in the appellant’s brief that were fabricated — stating explicitly that ‘fabricated cases, whether included by using A.I. or otherwise, are entirely inappropriate.’

An Eastern District of California immigration detention case turned into an AI accountability moment when the court identified two hallucinated case citations in the petitioner’s reply brief and directly addressed them at a January 2026 hearing, ordering counsel to inform all staff who assisted with the filing about the error.

A Texas Court of Appeals found that most of the case law cited by a self-represented litigant in a family law brief appeared to be fictitious — and expressly stated she ‘may have used an artificial intelligence tool’ in preparing her submissions. Despite the fabricated citations, the court addressed the merits on remand from the Texas Supreme Court.

A Kansas eviction appeal was decided against the tenant-appellant after her brief contained case citations that do not exist — identified by Westlaw editors as non-existent references preserved in the published opinion. The appellate court affirmed the default judgment of eviction while the AI citation misconduct was documented on the record.

A Delaware Court of Chancery judge noted that a petitioner’s brief contained several incorrect citations and warned that the use of generative AI to prepare court submissions without verifying accuracy is harmful to the legal system — even in estate and probate proceedings.