In 2025, AI-generated fabrications stopped being a curiosity and became a docket problem. Courts logged hundreds of filings containing citations to cases that do not exist. The lesson is not "avoid AI" — it is "never file what you cannot open."
The pattern behind the sanctions
Almost every sanctioned filing follows the same shape: a lawyer asks a general-purpose chatbot a legal question, the model returns confident prose with authoritative-looking citations, and nobody opens the cited source before it reaches a judge. The model was never reading a reporter — it was predicting plausible text.
The fabricated citation that triggers a show-cause order is rarely exotic. It is an ordinary cite — § 2025 WL —— — that simply has no document behind it.
Why "be careful" is not a control
Telling associates to "double-check the AI" is advice, not a control. Under deadline, verification is the first step that gets skipped. A durable fix changes the tool, not the willpower: the system should refuse to assert anything it cannot cite to a document you can open.
If it isn't in your files, the tool shouldn't say it.
That is the difference between a chatbot and citation-grounded AI. One generates language; the other answers from a specific record and shows you the page.
Christie cites every answer to your own files.
Get early access →A practical checklist
- Treat any AI output without an openable source as a draft prompt, not a fact.
- Prefer tools that ground answers in your matter and link each claim to a page.
- Keep a record of what was checked — an audit trail is your defense if a citation is ever questioned.
The sanctions of 2025 were not really about AI. They were about filing claims nobody verified. Citation-grounded tools make verification the default instead of the exception.
