The 15-minute recess: how a fabricated citation unravels at oral argument

July 5, 2026 · Every factual claim in this piece links to its source.

On May 20, 2026, an attorney stood before a Second Department panel in a routine sidewalk trip-and-fall appeal. Mid-argument, the judges started asking about specific cases in his brief — cases they could not find. He was offered a 15-minute recess to figure out where the citations came from. He declined, and told the court the authorities came from Westlaw, Lexis, or a book.

They came from none of those places. On June 23, in Landberg v. City of New York, 2026 NY Slip Op 03935 (PDF mirror), the Appellate Division sanctioned attorney Michael Sanders $8,000 and his firm $2,500, payable to the Lawyers' Fund for Client Protection — and left open whether the Attorney Grievance Committee takes it from here. The opinion, by Presiding Justice Hector LaSalle, is worth reading in full because it dissects, citation by citation, exactly what an unverified AI-assisted brief looks like from the bench.

Anatomy of the brief: three fakes, two misrepresentations

The court's inventory of the appellant's brief:

Xiang Fu Ji v. City of New York, 13 NY3d 791 — does not exist. Quoted twice for the claim that NYC's sidewalk-liability statute (Administrative Code § 7-210) is remedial and should be "construed liberally to effectuate its remedial purpose." There is a real Court of Appeals case with a similar name — Xiang Fu He v. Troon Mgt., 34 NY3d 167 — but it says no such thing. The Court of Appeals has actually held the opposite: § 7-210 is in derogation of common law and "must be strictly construed" (Vucetovic v. Epsom Downs, 10 NY3d 517). A fictitious quote, from a fictitious case, stating the reverse of the law.
Hausser v. Giunta, 88 AD3d 969 — does not exist at that citation. A real Hausser v. Giunta exists, but it's about a City of Long Beach ordinance, not § 7-210, and contains no version of the quoted language.
Lack v. Lack, 841 NYS2d 403 — no New York case by that name appears ever to have existed.
Campaign for Fiscal Equity v. State — real case, misrepresented. Cited for a rule about strictly construing exceptions to remedial statutes; the pinpoint page is actually about pleading standards under CPLR 3211(a)(7).
Rodgers v. City of New York — real case, misrepresented. Characterized as holding a two-year records search "can be sufficient" for prior-written-notice purposes; the court found Rodgers "said nothing remotely related" to that.

Nobody on the other side caught any of it. The court noted pointedly that counsel for both respondents — the City and the property owner — submitted briefs without alerting the court to the fabrications. The panel found them itself.

The part that turned $250 into $8,000

Weeks earlier, the same court sanctioned a pro se litigant $250 for one fabricated citation in Matter of Julien v. Arthur. Why 32× more here? Partly because Sanders is a trained attorney with Westlaw access. But the opinion dwells longest on what happened after the panel's questions: he declined the recess, attributed the cites to traditional research at the podium, and only in a written affirmation — filed under an order to show cause — admitted the fabricated cases likely came from "one of the free" AI tools he used and "negligently failed to verify," in violation of his own firm's AI policy. He wrote that he'd been "genuinely scared… afraid to even use the words 'AI.'"

The court's response: the lack of candor "demonstrated his professional immaturity, arrogance, and profound lack of respect for the judicial system." And it found the affirmation itself still incomplete — it addressed the three fake cases but never acknowledged the misrepresented real ones. The fear of saying "AI" out loud cost far more than the error itself.

What a verification pass would actually have caught

Honest scoping matters here, so let's be precise about which of the five problems a citation-existence check catches:

The three fabricated cases — instantly. Each fails a lookup against public court records. That check takes minutes, not the 15 the panel offered; run before filing, it's invisible; skipped, it becomes a published opinion with your name in the caption.
The two misrepresented real cases — no. Those citations exist and resolve to real decisions. Catching a mischaracterized holding still requires a lawyer reading the case. No automated existence check honestly claims otherwise — which is exactly why clearing the fakes mechanically matters: it frees the human hours for the part only humans can do.

The Second Department has now marked out its scale: $250 for a pro se litigant's single fake, $8,000-plus-firm-liability for an attorney's three fakes compounded by lack of candor — alongside the Third Department's $7,500 in Deutsche Bank v. LeTennier for 23 fabricated cases across five filings. The AI Hallucination Cases database stands at 1,725 decisions worldwide as of July 5, 2026. New York's appellate courts are no longer treating these as novelties; they are applying an established sanctions framework with escalating numbers.

Before you file — or before you argue against — run the brief through LegalCite. It checks every citation against public court records and flags cases that don't exist or don't match their case names. The three fabrications in Landberg are exactly the failure mode it catches. Free, no signup, and the document never leaves your browser: run it through Brief Check.

LegalCite verifies that citations exist in public court records (CourtListener) and match the case names attributed to them. It does not assess good-law status or verify quotations and holdings — the two misrepresented real cases in Landberg would require human review — and nothing here is legal advice. Citations to unpublished databases (Westlaw/Lexis) are reported as unverifiable, not fabricated. Sources: Landberg v. City of New York, 2026 NY Slip Op 03935 (App. Div. 2d Dept., June 23, 2026; PDF); NY Daily Record coverage (June 26, 2026); AI Hallucination Cases database (count verified July 5, 2026).