An attorney in Dubinin v. Papazian filed a brief with ten AI-fabricated citations, ending the case

In Dubinin v. Papazian, plaintiff's counsel Missiva Tilleli Khacer filed a response brief containing at least ten fabricated case citations and quotations attributed to nonexistent Eleventh Circuit opinions. The drafting had been delegated to New York attorney Nataliya Gavlin, whose legal assistant used generative AI to produce the brief. The U.S. District Court for the Southern District of Florida dismissed the case without prejudice, ordered Khacer to pay $4,030.90 in defendant's attorneys' fees, and referred all counsel to the Florida Bar and the court's Grievance Committee.

MTK International Law Group, PA · Incident Nov 21, 2025 · Indexed Jun 4, 2026 · 3 sources

A legal assistant fed a prompt to generative AI and the model conjured ten nonexistent cases and fabricated quotes that no attorney verified before filing.
What
In Dubinin v.
Incident date
Nov 21, 2025
Who
MTK International Law Group, PA
Failure mode
Hallucination
AI surface
Search / RAG
Severity
High

What happened

Plaintiff's counsel Missiva Tilleli Khacer filed a response to a motion to dismiss in Dubinin v. Papazian that contained at least ten non-existent case citations and fabricated quotations attributed to Eleventh Circuit opinions. Khacer had delegated the drafting to New York attorney Nataliya Gavlin, who was never admitted pro hac vice in the Southern District of Florida. Gavlin in turn had her legal assistant use generative AI to draft the brief. The defendant's reply exposed the fabricated citations, and the court confirmed through its own review that the cases and quotes did not exist. Judge Rodolfo A. Ruiz II struck the complaint, dismissed the case without prejudice, ordered Khacer to pay $4,030.90 in defendant's attorneys' fees, and referred both Khacer and Gavlin to the Florida Bar and the court's Grievance Committee.

What broke inside the model

Failure path · mode profile · Hallucination
  1. 01 · TriggerA user asks for a fact, a citation, or a figure.
  2. 02 · Model stepThe model writes a fluent, confident answer.
  3. 03 · Control gapNothing ties the claim back to a real source.
  4. 04 · FailureA fabricated fact ships as if it were verified.
  5. 05 · ConsequenceThe false claim reaches a customer, a court, or the public.

Confidence holds, and even spikes, as the claim detaches from any source.

Generative AI produced at least ten fictitious case citations and fabricated quotations that did not exist in any Eleventh Circuit opinion, a known failure mode where language models generate plausible-sounding but entirely invented legal authorities. The attorneys failed to verify the AI-generated content against actual case law before filing it with the court. Signing counsel delegated drafting through multiple layers of subordinates without supervising the output, allowing hallucinated citations to reach the court record undetected.

Public visibilityMedium
Regulatory exposureActive
Customer impactFew customers
Financial impactDisclosed
Time to disclosureDays
  1. Court FilingOrder Striking Complaint, Entering Sanctions, and Referring Counsel to the Florida Bar and Grievance Committee, Dubinin v. Papazian, No. 25-cv-23877-RARwebsitedc.s3.amazonaws.com
  2. PressAI IP Year in Review: AI Hallucinations in Court Filings and Orders: A 2025 Review of Sanctions Across the Courts and Rule Proposalssternekessler.com
  3. PressDubinin v. Papazian | Legal AI Governance Trackerlegalaigovernance.com
Permalinkhttps://failureindex.ai/failures/attorney-dubinin-v-papazian-filed-brief
CitationAI Failure Index. "An attorney in Dubinin v. Papazian filed a brief with ten AI-fabricated citations, ending the case" (FI-0130). Realm Labs. https://failureindex.ai/failures/attorney-dubinin-v-papazian-filed-brief (indexed Jun 4, 2026).
Share cardA branded image of this record for posts and slides.

Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0130. Full dataset at /data.

Note from Realm Labs, the Index steward

How Realm would have caught this

Controls for this failure mode
  • Prism
  • OmniGuard
  • AI Detection & Response (AIDR)

A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.