Federal Rule of Civil Procedure 11 requires that every filing be certified, by the signing attorney, as the product of a reasonable inquiry into the law and facts. Courts have not needed to write new rules to address AI-hallucinated citations — they have applied Rule 11 (and state-court analogues) directly, and the documented cases now give a clear, specific picture of how that has played out.
The baseline: Mata v. Avianca sets the template
In Mata v. Avianca, Inc., No. 22-cv-1461 (S.D.N.Y. 2023), Judge Castel found the Rule 11 reasonable-inquiry standard was violated when attorneys filed ChatGPT-generated citations, including a wholly fabricated case, without independent verification. The $5,000 sanction and the requirement to notify affected judges established the template most subsequent courts have followed: the AI tool is not itself the violation, the absence of verification before certifying the filing is.
Escalating sanctions for volume: Coomer v. Lindell
In Coomer v. Lindell (D. Colo.), Judge Nina Y. Wang found a filing from attorneys Christopher Kachouroff and Jennifer DeMaster contained nearly thirty defective citations and imposed $3,000 sanctions against each attorney under Rule 11, calling it “the least severe sanction adequate to deter and punish.” When the same attorneys repeated similar citation errors in a later filing, the court issued a show-cause order proposing a graduated $5,000 sanction — illustrating that Rule 11 sanctions in this area can and do escalate for repeat conduct.
Sanctions distributed by role, not applied uniformly: Wadsworth v. Walmart
In Wadsworth v. Walmart Inc. (D. Wyo. 2025), Judge Kelly Rankin's sanctions were calibrated to each individual's role in the failure: attorney Rudwin Ayala, who generated the fabricated citations using his firm's internal AI tool and did not verify them, was fined $3,000 and had his pro hac vice status revoked; attorneys T. Michael Morgan and Taly Goody, who signed the filing without independently verifying it, were each fined $1,000. Notably, the firm itself was not separately sanctioned, because it had already implemented a policy requiring independent verification of AI-generated content going forward — a detail worth noting for firms building their own policies.
When Rule 11 sanctions are declined: Michael Cohen
Not every documented case results in sanctions. In the Michael Cohen matter (S.D.N.Y.), Cohen had unknowingly provided his attorney, David M. Schwartz, with case citations generated by Google Bard, believing it was a standard search engine. Schwartz included them in a filing without independently verifying them. Judge Jesse Furman declined to impose sanctions, finding insufficient evidence of bad faith. This case is a useful counterpoint: it shows Rule 11 analysis in this area still turns on the traditional bad-faith and reasonable-inquiry factors, not on a strict-liability standard that automatically punishes any AI-related citation error.
Rule 11 outside federal court: state-court equivalents
State courts have applied their own equivalent rules with similar reasoning. Kruse v. Karlen (Mo. Ct. App. E.D. 2024) applied Missouri's frivolous-filing standard rather than federal Rule 11 directly, sanctioning a pro se appellant $10,000 for a brief where 22 of 24 citations were fabricated, reasoning that the filing forced the opposing party to expend unnecessary resources. The underlying logic mirrors federal Rule 11 even where the specific procedural vehicle differs.
The consistent pattern across every documented case
Reviewing the public record broadly (now past 1,750 documented cases per Damien Charlotin's database), a few consistent factors determine outcome severity: whether verification happened at all before filing, whether the attorney disclosed candidly once the issue was raised versus defending the fabrication, the volume of defective citations in a single filing, and whether it was a repeat occurrence. None of the factors turn on which specific AI tool produced the fabrication.
What this means for your own Rule 11 exposure
The practical takeaway is that Rule 11 compliance in the AI era requires the same reasonable-inquiry standard that has always existed — it just now specifically needs to account for the fact that an AI drafting tool's confident tone is not evidence of accuracy. A documented verification step, run consistently, is what satisfies the standard; it does not require avoiding AI tools altogether.
How Rule 11's safe harbor provision interacts with AI-citation errors
Rule 11(c)(2) includes a 21-day safe harbor: a party moving for sanctions must first serve the motion on opposing counsel and give them 21 days to withdraw or correct the offending filing before the motion can be filed with the court. This mechanism has mattered in several documented AI-hallucination cases, since an attorney who promptly withdraws or corrects a fabricated citation once notified, within the safe harbor window, may avoid a formal sanctions motion entirely — another concrete illustration of how much outcomes depend on the speed and candor of the response once an issue is flagged, not solely on the initial error.
The court's own inherent authority as a separate basis for sanctions
Beyond Rule 11 itself, courts also draw on their inherent authority to sanction conduct that abuses the judicial process, a broader and more flexible power than Rule 11's specific procedural requirements. Several documented AI-hallucination cases, including elements of Mata v. Avianca, invoke this inherent authority alongside or instead of Rule 11 specifically, which matters because inherent-authority sanctions are not subject to the same 21-day safe harbor procedure, giving courts a faster path to address egregious conduct.
Tracking how sanctions amounts have trended over time
Comparing sanctions amounts across the documented timeline suggests a rough upward trend as courts have become more familiar with the pattern and less inclined to treat it as a novel, sympathetic first-time mistake: earlier cases like Mata (2023) imposed $5,000, while more recent cases have both matched and, in repeat-offense scenarios like Coomer v. Lindell, proposed escalating beyond that figure. This is consistent with courts generally, over time, extending less benefit of the doubt to a now well-publicized risk.
A common question
Can a firm be sanctioned under Rule 11 even if it has an AI verification policy in place?
Based on the documented cases, having a policy in place appears to matter to outcomes (as in Wadsworth, where the firm itself avoided separate sanction partly because of its existing policy), but a policy that exists on paper without being followed in a specific instance will not protect the individual attorney who signed a defective filing. The policy needs to be both real and actually applied to the specific filing in question.