Repeat Offenders: The Lawyers Sanctioned Twice for AI-Hallucinated Citations

Part of: AI Citation Sanctions: What 49 Real Court Cases Show

The case that makes the pattern visible

Most discussion of AI hallucination sanctions treats each case as a discrete, isolated event: one attorney, one bad brief, one embarrassing court order. Cartagena v. Dixon, Blackburn, and T.A. Blackburn Law breaks that framing. The Southern District of New York’s July 2026 order didn’t just find at least 17 instances of quoted language that didn’t match the cited sources in the brief before it — it explicitly noted that the same attorney had already been sanctioned $5,000 in a prior, separate matter, Jakes v. Youngblood, for related conduct, before striking the Cartagena filing and referring the attorney to the Grievance Committee (citationsafe.com/sanctions-database/cartagena-v-dixon-blackburn-and-t-a-blackburn-law-2-20260710).

That’s a materially different fact pattern than a first offense. The attorney in Cartagena was not someone who had never been told the risk was real. A federal court had already imposed a five-figure-adjacent monetary sanction for the same underlying category of conduct. The second filing, with 17 instances of unsupported quoted language, happened anyway.

Why the court treated the prior sanction as relevant

Courts don’t have to reference an attorney’s disciplinary history when imposing a new sanction — the Cartagena order could have addressed the filing on its own terms without mentioning Jakes v. Youngblood at all. That the court chose to name the prior matter suggests the repeat nature of the conduct affected the court’s assessment of severity, and likely its decision to escalate all the way to a Grievance Committee referral rather than a lesser sanction. A single, isolated fabrication might plausibly be treated as a workflow failure or an oversight; a second instance, after a documented five-figure-range sanction for the same category of misconduct, is harder to characterize as anything other than a pattern.

What this implies about warnings as a deterrent

Citation Safe’s broader dataset includes several cases that resolve as a bare warning — the court identifies the problem, admonishes counsel, and moves on without a formal sanction (see, for example, Julia Rose v. Arts Bonita, Inc. and Jordan Slach v. City of Battle Ground, both warning-level outcomes). The Cartagena attorney’s history argues that even a much harsher intervention than a warning — an actual $5,000 sanction — did not reliably prevent recurrence. If a five-figure-adjacent monetary penalty in one matter doesn’t change an attorney’s filing workflow before the next matter, it is difficult to construct a confident argument that a bare warning, several tiers milder, would do better on its own.

This is the practical case for a mechanical, pre-filing verification step rather than relying on the deterrent effect of consequences after the fact: consequences that arrive after a filing has already gone to the court, however severe, cannot un-file the bad brief, and — per the one documented case where the record shows what happened next — do not reliably prevent the next one either.

The limits of what can honestly be said here

It would overstate the record to claim that repeat offenses are common across the tracked 49-case set. Citation Safe’s database does not publish a systematic cross-reference identifying every attorney who appears more than once, and this article is not in a position to assert a repeat-offense rate without that underlying analysis. What can be said honestly, and is fully supported by the public record, is that at least one such case exists, that the court treated it as material to the outcome, and that it directly illustrates a dynamic — warnings alone not changing behavior — that is otherwise easy to argue abstractly and hard to document concretely. Readers who want to know whether other attorneys in the tracked set have prior AI-citation sanctions elsewhere should check each individual attorney’s disciplinary and litigation history directly; that cross-referencing work has not been done systematically here or, as far as this article can confirm, by Citation Safe’s public database.

What this means for firm-level risk management

• A documented, real-world case exists where a five-figure-range sanction did not prevent recurrence. Treat “we’ve already had one incident and dealt with it” as insufficient assurance against a second one, absent a structural process change.
• Courts appear willing to weigh disciplinary history when deciding how far to escalate. A second incident is not evaluated in a vacuum — the Cartagena order shows a court explicitly considering the Jakes v. Youngblood sanction when deciding to refer the attorney to the Grievance Committee.
• The honest data limitation matters. No systematic repeat-offense rate exists in the public record right now; firms should not assume repeat sanctions are either rare or common without better underlying data than currently exists.
• Process changes, not personnel warnings, are the documented answer. A firm response limited to “make sure this doesn’t happen again” without a mechanical verification gate is, per the one fully-documented case available, not obviously sufficient.

Related reading

• AI Citation Sanctions: What 49 Real Court Cases Show — the pillar this spoke expands on.
• Bar Referral: What Actually Happens After a Court Refers a Lawyer to Disciplinary Counsel — what happens procedurally after the Cartagena referral.
• citationsafe.com/sanctions-database
• citationsafe.com/verify

Sources

• Citation Safe, Cartagena v. Dixon, Blackburn, and T.A. Blackburn Law — source of the Jakes v. Youngblood prior-sanction reference.
• Citation Safe, Public AI Legal Sanction Case Database, accessed 2026-07-16.

Citation Safe is a verification workflow tool, not legal advice. No attorney-client relationship is created. Human review is always required. This article documents one specific, publicly-recorded repeat-sanction case and does not assert a systematic repeat-offense rate across the tracked database, which has not been independently computed.

Check a brief before you file it →