The AI Learning Hub Journal

The Sanctions Pattern

The recurring anatomy of an AI-citation failurea pattern, not an incident — no matter, forum, outcome or figure is described here1Time pressurea deadline, a late instruction, a long list2The tool produces plausible authoritieswell-formed, on point, confidently stated3No independent verificationthe shape of the answer is taken as the check4The work is filed or senta signature attaches a person to the content5Someone goes looking for the authorityan opponent, a judge or the client cannot find it6Consequences follow the person and the firmprofessional, reputational and commercial at onceWHERE THE CHAIN BREAKSEvery other step here is ordinarypractice. Deadlines are normal.Fluent text is what these toolsproduce. Filing is the job.Only one step is missing, and itis the one that stops all of it:OPEN THE ACTUAL SOURCEAND READ IT BEFORE FILINGNot the tool's summary of it.Not a second tool's agreement.The source itself, read by you.Insert that one step and therest of the chain cannot form.One habit, applied every timeThe pattern is mundane at every step except one — which is why it repeats, and why a single habit interrupts it
Every step but one is ordinary professional life — the pattern survives because the missing step is the only unusual thing about it

What Has Actually Been Happening

Since generative tools became widely available, a consistent pattern has emerged across multiple jurisdictions: filings submitted containing citations to authority that does not exist, discovered either by opposing counsel unable to locate the cases or by the court itself. The most widely reported early instance is Mata v. Avianca in the Southern District of New York in 2023, where a filing cited multiple invented cases produced by a chatbot and the attorneys were sanctioned — but it was the first famous example of the pattern, not an aberration. Since then the responses have included orders to show cause, monetary sanctions, referrals to disciplinary bodies, adverse findings recorded in published judgments, striking of the offending filing, and in some instances costs awarded against the responsible practitioner personally. Reported incidents have involved practitioners at a wide range of experience levels and firm sizes, and they have continued well after the risk became widely publicised. This is a documented and recurring phenomenon, not a single memorable episode.

  • Mata v. Avianca (S.D.N.Y. 2023) is the canonical example — invented citations, sanctioned attorneys, worldwide coverage
  • Consistent pattern across jurisdictions: filings citing authority that does not exist
  • Consequences have included show-cause orders, monetary sanctions, disciplinary referrals, and adverse published findings
  • Incidents span experience levels and firm sizes — seniority has not been protective
  • The pattern has continued after widespread publicity, which tells you awareness alone is not a control

What Aggravates the Consequence

The reported outcomes suggest the initial error is rarely what determines severity. What aggravates it is the response: failing to check when opposing counsel raises the issue, defending the citations rather than verifying them, offering explanations that shift as facts emerge, and denying AI use before conceding it. Courts have generally distinguished between a practitioner who made a verification failure and admitted it promptly, and one who compounded it. The duty of candour engages the moment a doubt is raised, and by then the situation is recoverable if handled directly. The related lesson is that a firm needs a rehearsed response for the day this happens — check immediately, correct on the record, disclose plainly — because improvising it under pressure has gone badly for people.

  • Severity tracks the response more than the original error
  • Defending unverified citations rather than checking them has consistently made outcomes worse
  • Candour obligations engage as soon as a doubt is raised, not when the fabrication is proven
  • Have a rehearsed response: verify at once, correct on the record, disclose plainly

The Institutional Response

Courts have not only sanctioned individuals. Judges in a number of jurisdictions have issued standing orders or practice directions addressing AI use in filings — some requiring disclosure of whether generative AI was used, some requiring certification that citations have been verified by a human, some restricting use in particular categories of filing. These requirements are inconsistent between courts and between judges within the same court, and they are still evolving. Bars and regulators in several jurisdictions have also issued guidance. The practical consequence is that "what do I have to disclose?" now has a per-forum answer, and a practitioner appearing in an unfamiliar court has a new item on the pre-filing checklist.

  • Standing orders and practice directions on AI use exist in a number of courts and vary widely
  • Requirements range from disclosure of use, to human-verification certification, to category restrictions
  • They differ between courts and between judges in the same court, and they are still changing
  • Checking the forum's current requirements belongs on the pre-filing checklist, every time

Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.