Braica v. Frankowski
U.S. District Court, District of Connecticut · D. Conn. · Connecticut bar guidance
Conduct
Pro se plaintiff filed Response and Sur-Reply riddled with fabricated cases, fabricated quotes, and misstatements of law; OSC response admitted using AI websites without independent verification.
Consequence
Strike of Response and Sur-Reply without leave to amend; admonition; warning that future AI hallucinations will trigger more stringent sanctions, including dismissal of the case with prejudice.
Lesson
Strike-plus-admonition is the calibrated mid-tier sanction: harsher than warning-only (no monetary), milder than fees-and-fees. Chambers patterns absent local AI rules can themselves do the escalation work.
Verified May 10, 2026
- Citation
- Braica v. Frankowski, No. 3:24-cv-01709-VDO, Memorandum & Order (D. Conn. Dec. 15, 2025) (Oliver, J.), ECF No. 35
- Decided
- December 15, 2025
Summary
Pro se plaintiff Anthony Braica, a probationer, brought Fourth, Eighth, and First Amendment civil-rights claims (plus intentional and negligent infliction of emotional distress) against probation officer Tom Frankowski for an alleged sexual assault during a search to detect drug-test falsification devices. Defendant filed a partial motion to dismiss the Eighth Amendment and negligent-infliction claims. Plaintiff's Response (ECF No. 30) and subsequent improper Sur-Reply (ECF No. 32, filed without leave) were strewn with AI-generated hallucinations across three categories: entirely fabricated cases (State v. Kelleher and Sheppard v. Roberts, citations leading to unrelated decisions), fabricated quotations attributed to real cases (Martin v. Brady, State v. Moore), and misstatements of law (Solem v. Helm and Graham v. Florida cited for incarceration-Eighth-Amendment law as applied to probationers; Rochin v. California cited for Eighth Amendment when it analyzed Thirteenth/Fourteenth; Steffel v. Thompson cited for Eighth when it concerned First/Fourteenth; United States v. Jones cited for GPS attachment to a person when it concerned attachment to a vehicle; Quon v. Arch Wireless cited without disclosing it was reversed by the Supreme Court; Hamilton v. Lajoie cited for a proposition the case ultimately rejected). The Court issued an Order to Show Cause (ECF No. 33, Oct. 3, 2025); Plaintiff's Response to the OSC (ECF No. 34) admitted to using 'publicly available websites' and accepting their output without independent verification, but the Sur-Reply continued to contain new fabricated cases and misstatements.
- AI tool:
- Unspecified generative AI
- Sanction amount:
- Admonition + strike of Plaintiff's Response (ECF No. 30) and Sur-Reply (ECF No. 32) from the record without leave to amend; future sanctions (including dismissal of the case with prejudice) reserved for any further AI-generated hallucinations in any subsequent filing
What sanction did the court impose?
On December 15, 2025, Judge Vernon D. Oliver issued a 12-page Memorandum & Order applying the two-factor Andre v. Warden framework (verification and inquiry; candor and correction) and finding Plaintiff failed both factors. The Court found 'subjective bad faith' satisfying the sua sponte Rule 11 standard, citing Park v. Kim, 91 F.4th 610, 615 (2d Cir. 2024) ('At the very least, the duties imposed by Rule 11 require that [litigants] read, and thereby confirm the existence and validity of, the legal authorities on which they rely.') and Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 464 (S.D.N.Y. 2023). The Court ADMONISHED Plaintiff for repeatedly violating Rule 11 and STRUCK the Response and Sur-Reply from the record without leave to amend. The Court cautioned: 'it will impose more stringent Rule 11 sanctions, including dismissal of the case with prejudice, upon identifying AI-generated hallucinations in any future filings.' No monetary sanctions were imposed. The Court cited Andre v. Warden, No. 3:24-CV-01295 (VDO), 2025 WL 3281732 (D. Conn. Nov. 25, 2025) (the chambers framework precedent), Benjamin v. Costco Wholesale Corp., 779 F. Supp. 3d 341, 350 (E.D.N.Y. 2025), Ramirez v. Humala, No. 24-CV-242, 2025 WL 1384161 (E.D.N.Y. May 13, 2025), Kaur v. Desso, No. 9:25-CV-726, 2025 WL 1895859 (N.D.N.Y. July 9, 2025), and Moales v. Land Rover Cherry Hill, No. 3:25-CV-544 (VDO), 2025 WL 1249616, at *3 (D. Conn. Apr. 30, 2025) for the AI-as-helpful-tool framing.
Why does Braica v. Frankowski matter for law firms using AI?
The Braica order is the second chambers application of the Andre v. Warden two-factor framework in D. Conn., and the first to add the explicit dismissal-with-prejudice threat for repeat conduct. The combination of (a) actually striking the offending pleadings, (b) finding subjective bad faith under the heightened sua sponte Rule 11 standard, and (c) reserving dismissal with prejudice for the next occurrence places this order at a sanction tier above the warning-only template most other 2025-2026 pro se AI-citation orders apply.
The Andre framework imported here is itself imported from Mattox v. Production Innovations Research (E.D. Okla. Oct. 22, 2025). The cross-circuit borrowing pattern (Oklahoma to Connecticut chambers via the Andre opinion) is the same pattern visible in the Cartwright orders (Romero v. Goldman Sachs, S.D.N.Y. to W.D. Wash.) and the Kiel order in Baker v. Rastelli (Mullins v. Duquesne, W.D. Pa. to D.N.J.). For partners tracking judicial AI policy, the operative case law in 2026 is a mesh of cross-cited chambers frameworks rather than a circuit-bound common law.
The three-category hallucination taxonomy (fabricated cases / fabricated quotations / misstatements of law) is the most analytically useful piece of this opinion for defense counsel drafting AI-citation sanctions briefs. The first two categories are objectively verifiable; the third is judgment-laden. Judge Oliver’s framework for the third category, looking to prevalence of other AI signals in the filings and egregiousness of the misstatement, gives a tractable test for distinguishing AI-driven misstatements from genuine pro se misunderstandings. Briefs that argue all three categories together with this taxonomy will read as more disciplined than briefs that argue “fake cases” alone.
The candor factor here also has teeth. Plaintiff apologized but the Sur-Reply continued to introduce new fabricated cases, and the OSC response stopped short of identifying generative AI specifically (referring instead to “publicly available websites”). The Court treated this concealment-by-paraphrase as aggravating, citing Benjamin v. Costco Wholesale Corp. for the proposition that “regret and apologies are not necessarily enough to avoid the imposition of sanctions for the submission of non-existent legal authority.” Defense counsel facing similar pro se conduct should preserve the candor-and-correction prong on the record from the OSC response forward, treating each subsequent filing as a fresh test of the litigant’s commitment to verification.
Implications for your firm
Operational steps a firm reading this case may wish to consider documenting. Strategic and rule-application calls belong to your firm's attorneys.
- Document the Vernon D. Oliver chambers framework imported from Andre v. Warden into Braica: two factors, (1) verification and inquiry, (2) candor and correction. The framework itself is adapted from Mattox v. Production Innovations Research (E.D. Okla. Oct. 22, 2025), confirming the cross-circuit borrowing pattern in 2025-2026 chambers AI policies.
- Track the Andre-adopted three-category hallucination taxonomy: (1) fabricated cases (nonexistent names, invented citations, or real citations pointing to unrelated decisions); (2) fabricated quotations from actual cases; (3) misstatements of law (representations of legal rules or holdings that are inaccurate, incomplete, or unsupported). Defense counsel drafting AI-citation sanctions briefs should adopt this three-part taxonomy when categorizing offending citations.
- Note the sanction calibration: strike + admonition + dismissal-with-prejudice threat is one step below monetary sanction but materially stronger than the W.D. Wash. or D.N.J. warning-only template. The Vernon Oliver chambers pattern (Braica, Andre, Roark v. American Airlines) demonstrates a chambers-level escalation posture absent local AI standing orders.
- Document the inferential framework for the third hallucination category (misstatements of law): Judge Oliver looks to (1) prevalence of other obvious or admitted AI-generated content in the filings, and (2) egregiousness of the misstatement. This rebuts the 'pro se misunderstanding, not AI' defense that pro se filers might raise on a single misstated citation.