⚡ Case Digest
Butler v. Fidelity Management & Research Company — S.D.N.Y., April 30, 2026
A magistrate judge in SDNY reviewed a pro se plaintiff’s procedural letter requesting administrative case docketing corrections and found it contained five case citations that did not support their cited propositions, three fabricated quotations attributed to real cases that did not contain the quoted language, and one citation to a completely nonexistent judicial opinion. The judge attributed these to generative AI use and warned that presenting false citations is sanctionable.
Why it matters: Even a routine procedural letter to the court must be accurate — AI hallucinations in any court submission, not just formal briefs, can result in sanctions.
Category: AI Hallucination & Sanctions | Jurisdiction: USA (New York) | Read time: 6 min
Case at a Glance
| Full Citation | Butler v. Fidelity Management & Research Company LLC (FMR) et al., Case No. 25-CV-09337 (RA) (RFT) (S.D.N.Y.), April 30, 2026 |
| Court | United States District Court, Southern District of New York (Magistrate Judge Tarnofsky) |
| Date | April 30, 2026 |
| AI Tool / Issue | Procedural letter contained 5 cases that did not support cited propositions, 3 fabricated quotations from real cases, and 1 nonexistent opinion — all in a routine administrative letter |
| Outcome | Letter’s substantive request denied; AI hallucination warning issued; court puts plaintiff on notice that false citations in any submission are sanctionable |
Background
Pro se plaintiff Sana Hazina Butler filed a letter to the court requesting administrative corrections to how her removed state-court complaint was docketed — she wanted the complaint to appear as a standalone entry rather than as an exhibit to the notice of removal. In support of her request, Butler cited six cases, three of which she quoted directly.
The magistrate judge reviewed each citation. Granny Goose Foods v. Bhd. of Teamsters, 415 U.S. 423 (1974), exists and involves removal but says nothing about whether a removed complaint may be filed as an exhibit; none of the quoted language appears in the opinion. Rong Zhu v. BNP Paribas, 997 F. Supp. 2d 242 (S.D.N.Y. 2014), does not exist at all. Parisie v. Greer, 705 F.2d 882 (7th Cir. 1983), exists but has nothing to do with removal. Azer v. Connell, 306 F.3d 930 (9th Cir. 2002), exists but has nothing to do with removal and the quoted language does not appear in it. Maersk, Inc. v. Neewra, Inc., 554 F. Supp. 2d 424 (S.D.N.Y. 2008), exists but has nothing to do with removal and the quoted language does not appear in it. Whitaker v. Am. Telecasting, Inc., 261 F.3d 196 (2d Cir. 2001), exists and involves removal but says nothing about whether a removed complaint may be filed as an exhibit.
The judge denied the request on its merits (noting the standard SDNY practice of docketing removed complaints as exhibits is fine), then separately addressed the citation problems.
The AI Issue
The court stated that “I believe that these issues are due to Plaintiff’s use of generative AI tools,” citing the pattern of real cases with wrong propositions, fabricated quotations attributed to real cases, and a completely nonexistent citation. The court then quoted the SDNY standard warning from Romero v. Goldman Sachs Bank USA (2025): “citations, quotations, and holdings that have been generated by generative artificial intelligence may be hallucinations” and “the presentation of false citations, quotations, and holdings by a party to the Court is sanctionable conduct.”
What the Court Decided
- Even a routine procedural letter to the court — not a formal motion or brief — must comply with the duty of accuracy under Rule 11 and the court’s inherent authority to sanction misrepresentations.
- Fabricated quotations attributed to real cases that do not contain the quoted language are misrepresentations of the law just as much as citations to nonexistent cases.
- AI hallucinations in court submissions — whether in briefs, motions, or letters — are inferable from citation patterns and are sanctionable conduct regardless of subjective intent.
- The court put Butler on notice that future submissions containing false citations, quotations, or holdings will be subject to sanctions proceedings.
- The Second Circuit’s Park v. Kim (2024) principle was cited: the “attempt to pass off fake citations as real is a fraud on the court.”
“The presentation of false citations, quotations, and holdings by a party to the Court is sanctionable conduct.”
— Citing Romero v. Goldman Sachs Bank USA; adopted by Magistrate Judge Tarnofsky, SDNY, April 30, 2026
The India Angle
Indian Law Equivalent
Indian procedural practice includes letters, applications, and miscellaneous petitions filed with courts that are less formal than plaints or written statements. Under the Supreme Court Rules, 2013, all applications and letters filed in proceedings must comply with the general duty of accuracy. The Contempt of Courts Act, 1971, Section 2(c) definition of criminal contempt covers any act that “scandalizes or tends to scandalise” the court — including presenting fabricated quotations attributed to judgments that do not contain them, which constitutes a direct attack on the integrity of existing judicial opinions.
Bar Council Rules
BCI Rules, Chapter II, Rule 22 applies to all representations made in proceedings, not only to formal pleadings. A letter or application to the court containing fabricated case quotations violates this rule. Rule 9 prohibiting acts that mislead the court applies with equal force to informal and formal submissions.
Practical Advice for Indian Advocates
- Apply the same citation verification standards to every communication to the court — letters, notes, applications, and informal submissions carry the same duty of accuracy as formal briefs.
- Never use AI-generated quotations from case judgments without reading the actual passage in the original judgment to confirm the words appear verbatim — AI tools frequently fabricate convincing but non-existent judicial language.
- Be especially careful with quotations attributed to well-known cases: AI tools frequently insert hallucinated language into famous cases because the model has seen those cases referenced frequently in training data.
Quick Takeaways
- AI hallucinations in procedural letters carry the same sanctions risk as hallucinations in formal briefs.
- Fabricated quotations from real cases are independently sanctionable misrepresentations of law.
- SDNY has standardized AI-hallucination warning language that is now routinely applied in all such cases.
Deep Dive: AI Hallucinations in Procedural Letters — An Underappreciated Risk
The Butler case expands the AI hallucination sanctions risk to a context that many practitioners may not have considered: the routine procedural letter to the court. Lawyers and pro se litigants often approach these letters with less rigor than formal briefs, viewing them as administrative communications rather than legal documents requiring the same verification standards. Butler establishes clearly that the duty of accuracy applies to every submission to the court, however informal its purpose.
The specific citation errors in Butler also illustrate the full range of AI hallucination types in a single document: one completely nonexistent case; several real cases cited for completely wrong propositions; and fabricated quotations inserted into real opinions that never contained them. This comprehensive variety in a single letter suggests that Butler did not write any of the citations herself — she fed the request into a generative AI tool and included the output verbatim without reading any of the underlying cases.
The fabricated-quotation issue deserves particular attention. When a party cites Granny Goose Foods v. Brotherhood of Teamsters for a proposition about removed complaints and includes a quotation in quotation marks that does not appear in the opinion, they are not merely misusing a case — they are fabricating language and attributing it to the Supreme Court. This is a qualitatively different kind of deception from a wrong citation: it puts false words in the mouths of real judges, creating a false record of what courts have held. Courts treat this as particularly serious, as reflected in the Second Circuit’s Park v. Kim “fraud on the court” formulation.
For Indian advocates, the lesson extends to all forms of court communication. In Indian practice, interlocutory applications, chamber applications, letters to the Registrar, and informal applications in chambers all constitute communications to the court. Any of these documents that contains AI-generated content — including fabricated case quotations — creates the same liability as a formal written submission. The duty of accuracy is not a function of the document’s length or formality; it is a function of its destination.