Do courts actually sanction fabricated AI citations when detected?

Damien Charlotin's AI Hallucination Cases database compiles court filings where fabricated or AI-suspected citations surfaced, and the five entries in this excerpt show detection running well ahead of punishment. A self-represented appellant's brief had "multiple hallucinated citations"; the court dismissed the appeal, but for noncompliance with briefing rules, not for the citations themselves. A pro se brief cited three nonexistent Kentucky cases (Hunt v. Smith, Bargo v. Bargo, Roberts v. Hensley) plus several real cases that "did not support the propositions for which they were cited"; the appellate court treated this as "a substantive violation of the appellate rules," limited its review to manifest injustice, and affirmed — again, no separate sanction for the fabrication itself. Defense counsel admitted "using a Westlaw AI tool to identify cases and copying propositions without verifying the authorities or citations," and the court found a Rule 11 violation in citing one fictitious case — yet affirmed the decision not to issue a show-cause order or impose sanctions, since "Rule 11 sanctions are discretionary." A UK tribunal judge found two cited Upper Tribunal decisions "appeared to be AI hallucinations" and simply fell back on current policy documents instead, with no professional sanction or penalty imposed. And in the fifth case, what looked like hallucinated citations — reporter numbers pointing to unrelated cases — turned out to be, in the court's own finding, "inadvertent typographical errors," with sanctions denied for lack of bad faith.

The courts' own language shows what actually drives the sanction decision: discretion ("Rule 11 sanctions are discretionary"), demonstrated harm ("no sufficiently demonstrated burden requiring further proceedings"), and intent ("no evidence of bad faith, deception, or frivolous legal argument," with the genuine cases "readily locatable"). Detecting a fabricated or wrong citation establishes that an error occurred; it does not by itself establish the fault or damage that triggers a penalty. Procedural noncompliance (the dismissed appeal) and a substantive-rule violation (the manifest-injustice case) did the work of deciding the outcome in two of the five cases, with the hallucination finding treated as incidental rather than central to the result.

This complicates how to read the aggregate count in Do small law firms misuse AI more often than large ones?: that note tallies who gets caught, but these five cases show that being caught rarely produces a matching sanction, and in one instance the citation wasn't an AI hallucination at all, just a typo mistaken for one. Where How often do legal AI tools actually hallucinate citations? measures how often sanctioned legal-AI tools fabricate citations before a filing is ever made, this excerpt picks up the story afterward — once a suspect citation reaches a judge, even a confirmed Rule 11 violation doesn't guarantee consequences beyond an admonition.

This is five entries, not a disclosed representative sample of the full database, and the excerpt doesn't say how the database curates or labels cases, so no sanction rate can be read off these five alone. It also only shows what happens once a citation is flagged, not whether courts under-detect hallucinations that go unflagged. The cautious implication: a rising case count in a hallucination database measures detection events, not confirmed AI misconduct or consistent judicial punishment, and both over-counting (typos mistaken for AI) and under-sanctioning (confirmed violations left unpunished) are live possibilities behind any single tally drawn from it.