Park v. Kim AI Sanctions: The Second Circuit Pro Se Citation Case
The second case in the canonical sequence is an appeal. The first was Mata v. Avianca, a district-court motion in the Southern District of New York that produced the first published sanctions order over AI-generated citations. Park v. Kim sits one rung up the federal hierarchy. The Second Circuit heard the appeal. The Second Circuit read the brief. The Second Circuit referred the attorney who filed it to the court's grievance panel.
The underlying mechanism is the same in both cases. The citations did not resolve to opinions in the reporters named. The procedural posture is different. The consequences are different. The lesson for a pre-filing workflow is identical.
The matter, in one paragraph
Park v. Kim, No. 22-1869, reached the Second Circuit on appeal from a district-court ruling. The attorney filed a brief on behalf of the client. The brief cited authority in support of the position. The panel could not locate certain of those citations in the reporters the brief named. The court entered an order directing the attorney to address the citations. The attorney's response did not satisfy the panel. The court's decision referred the attorney to the Second Circuit's grievance panel for consideration of discipline. Press coverage of the order noted that the brief appeared to contain citations generated by a large language model and submitted without verification against any reporter.
The procedural posture: why the appellate setting changed what was visible
At the trial level, citations move past several readers in compressed time. The associate cite-checks. The partner skims. The opposing party files a response. The judge reads. The judge's clerk reads. Errors of citation existence are caught at one of those stations, or they are not caught at all. The schedule and the volume of trial motion practice put pressure on the verification step.
At the appellate level, the posture is different. The brief is the principal vehicle. The panel reads it in full. The panel is not triaging the brief against twelve other motions on the same docket day. The panel is reading the brief as the basis on which the appeal will be decided. A citation that does not resolve to an opinion in the reporter named is, in that setting, conspicuous.
The Park v. Kim order frames the duty at the appellate level in familiar terms. An attorney owes the court candor. An attorney owes the court a brief whose authorities can be located in their reported form under Model Rule 3.3. The court's reasoning treats those duties as preexisting and unchanged. The AI tool is not the subject of the court's censure. The unverified filing is.
What the court said the duty was
The Second Circuit's order describes the duty in operational terms. An attorney who signs a brief represents to the court that the citations in the brief refer to real authority. That representation has been the bar standard for as long as briefs have carried signatures. The introduction of a generative tool into the drafting workflow does not alter the representation. The attorney who submits the brief remains responsible for confirming that the authority cited can be located in the reporter.
The court did not announce a new rule. The court applied the rule already on the books to a new factual pattern. The pattern is the one Mata v. Avianca made visible. A model produced a citation. The citation looked like the kind of citation a research platform would return. The brief was filed without the citation being checked against any reporter. The reporter, when checked, did not contain the opinion.
The procedural consequence was a referral. The Second Circuit maintains a grievance panel that considers attorney conduct within the court's bar. The referral directs the panel to consider whether discipline is warranted. The grievance process is separate from the appeal itself and separate from any state-bar discipline that may follow.
Why the appellate failure mode was harder to absorb
A monetary sanction at the trial level is, in one sense, a closed loop. The court issues the order. The firm pays. The matter file records the resolution. The sanction is visible on the docket but it is also bounded. The cost is known.
A referral to a grievance panel is not bounded in the same way. The panel's consideration is a separate proceeding. The reputational surface is wider. Future filings before the same court arrive in a posture where the panel's consideration is part of the record. The downstream effects on the attorney's practice are not easily predicted at the moment of the referral.
For the firm, the implication is straightforward. The cost of an unverified appellate citation is not measured by the size of a single sanctions order. The cost is measured by the long tail of a disciplinary inquiry, by the firm's standing before the court, and by the malpractice carrier's view of the firm's process. None of those costs are easy to convert into a number on the night the brief is signed.
What a pre-filing scanner returns on these inputs
The reply brief in Park v. Kim cited one case the Second Circuit could not find: Matter of Bourguignon v. Coordinated Behavioral Health Servs., Inc., 114 A.D.3d 947 (3d Dep’t 2014). We ran that citation, as it appeared in the brief, against CourtListener’s corpus.
The result is the same failure mode as Mata, in its subtler form. The address 114 A.D.3d 947 is real — an opinion is published there. It is simply not Matter of Bourguignon, which, as both the attorney and the Second Circuit confirmed, does not exist. The citation has the surface form of authority. The resolution finds that the named case is not at the address it claims.
The verdict is structured, and it would have been available to anyone reviewing the appellate brief before it was filed. The decision to file would have been made against a flagged citation, not against the unmarked text of the draft.
The mechanism is not subtle. A citation either resolves to an opinion in the reporter or it does not. A scanner that performs the resolution returns a verdict on each attempt. When the input is a hallucinated citation, the verdict is negative. That is the test Park v. Kim failed, and that the audit layer is built to run.
The sequence so far
Every sanctions case in the canonical sequence has the same mechanism at its center. The mechanism is a citation that did not exist, a brief that included it, and a filing step that did not run a verification against the reporter. The procedural posture varies. The court varies. The sanction varies. The mechanism does not.
The institutional response has converged. Federal districts have entered standing orders on AI disclosure and citation verification. State bar associations have published opinions framing the duty of technology competence to include verification of AI-generated work. Appellate panels, as Park v. Kim shows, are willing to refer attorneys to disciplinary review when the verification step did not run. The question for the firm is no longer whether the audit step is required. The question is what artifact the firm produces to show the audit ran.
The Verification Certificate is that artifact. The certificate lists the citations checked, the verification status for each, the source records consulted, a SHA-256 hash of the certificate content, a public token URL, and a timestamp. If the brief in Park v. Kim had been filed with such a certificate alongside it, two outcomes change. The brief is not filed in the form it took, because the certificate would have shown unresolved citations. Or, if the brief were filed anyway, the record would carry an affirmative decision to file unverified authority rather than an absence of verification.
The audit layer is the operational answer to a failure mode the profession has now observed enough times to name. Veritas runs the layer. The Filing Risk Scanner extracts the citations, resolves them against reporter and public-record sources, classifies each by verification status, and produces the certificate. The partner signs the brief after reading the certificate, not before.