A court order shows the cost of an unchecked citation

Start here

Asking an assistant to find supporting case law or a relevant statute is a reasonable first step in preparing a legal or financial question; the risk is treating the answer as the citation itself rather than a lead to verify. A federal court's own docket documents, in detail, what happens when that verification step is skipped.

What the documents say

In an Order to Show Cause dated 4 May 2023, Judge P. Kevin Castel of the Southern District of New York documented that a brief filed in the case cited six judicial decisions the court could not locate; the order lays out, case by case, that the decisions carried real-sounding docket numbers and citation formats attached to opinions that do not exist, along with internal quotes and case citations inside those opinions that were themselves invented. The order directed the attorneys to show cause why they should not be sanctioned. The court's follow-up Opinion and Order on Sanctions, dated 22 June 2023, states directly that the attorneys 'submitted non-existent judicial opinions with fake quotes and citations created by the artificial intelligence tool ChatGPT, then continued to stand by the fake opinions after judicial orders called their existence into question.' The order found bad faith specifically in that continued reliance after being questioned, not in the initial research error alone, and imposed a $5,000 penalty along with a requirement to notify the real judges falsely named as authors of the fake opinions. Separately, OpenAI's usage policies state that tailored advice requiring a license, including legal advice, should not be given 'without appropriate involvement by a licensed professional,' a policy line addressing exactly the gap this order describes.

Check this

Treat any case name, statute, or citation an assistant gives you as a lead, not a fact, until you have opened it in a source that indexes real case law or statutes and confirmed the quoted language appears there. The court's order shows the specific failure mode to watch for: a citation that looks correctly formatted, with a plausible docket number and court, can still refer to nothing.

What holds and what fails

The record supports a narrow, useful lesson rather than a broad one: an assistant can generate a citation-shaped answer without any of its citations being real, and that failure was not caught by the format looking right. It does not establish, and the order does not claim, that every legal or financial answer from an assistant is unreliable, only that unverified citations reached a federal filing in this case, with the consequences the order describes. This is an editorial boundary drawn from a single, well-documented record, not a survey of outcomes across cases.

• Verify every case citation or statute in a source that indexes primary law.
• Ask an assistant to help frame the legal or financial question, not answer it.
• Bring a verified question, not an unverified answer, to a licensed professional.

The order's detail is specific for a reason: it shows a fabricated citation can pass a casual read, and only checking the primary source catches it.