Jarrus v. Governor of Michigan, No. 25-cv-11168 (E.D. Mich. 2025)

Jarrus et al v. Governor of Michigan et al Doc. 176

UNITED STATES DISTRICT COURT EASTERN DISTRICT OF MICHIGAN

SOUTHERN DIVISION

MICHAEL D. JARRUS and LINDA JARROUS,

Plaintiffs,

v. GOVERNOR OF MICHIGAN, et al.,

Defendants.

___________________________ /

IMPOSING SANCTIONS FOR USE OF CHAT GPT (ECF Nos. 170,

*171)*

This order is issued upon receipt of Plaintiff’s response to an order

to show cause for use of so-called artificial intelligence (“AI”) in drafting their objections to an order issued by the Magistrate Judge. ECF No. 170 (objections); ECF Nos. 174 (response), 172 (declaration of Plaintiff Michael Jarrus), 173 (declaration of Plaintiff Linda Jarrous).

Although Plaintiffs were warned by Magistrate Judge Patti of the

dangers of the use of generative AI, Plaintiffs nonetheless appeared to make use of such a tool without checking the results. See ECF No. 170, PageID.3194 (including a clearly AI-generated phrase, “Here’s the

Dockets.Justia.com

Case No. 25-cv-11168

F. Kay Behm U.S. District Judge

Anthony P. Patti U.S. Magistrate Judge

revised Paragraph 2, . . .”). Even on first review, their briefing

appeared to be nothing more than a copy-paste from a chatbot-style

generative AI tool. Ironically, Plaintiffs objected to the Magistrate

Judge’s warning about AI use in the same document that they relied on

AI-generated text.

Upon more careful review of the objections, the court identified a

series of citations in Plaintiffs’ brief that were false citations (with real

cases, but with explanations that did not reflect the case cited).

For example:

• Plaintiffs cited Krupski v. Costa Crociere S. p. A., 560 U.S.

538 (2010), for the proposition that if a plaintiff seeks to amend their complaint and an added defendant “knew or should have known” that the action would have been brought against them but for a mistake (i.e. that the amendment would “relate back”), that “amendment is not discretionary - it is mandatory.” Krupski does not say that the court must permit amendment under that circumstance. Rule 15(a), not Rule 15(c), sets forth the circumstances in which a party may amend its pleading before trial.

• Plaintiffs cited District of Columbia v. Heller, 554 U.S. 570

(2008), for the proposition that “the loss of Second Amendment deprivations, even for minimal periods, constitutes irreparable injury.” Heller does not say that a temporary loss of Second Amendment rights is irreparable harm; Plaintiffs are extending a statement from Elrod v. Burns, 427 U.S. 347, 373 (1976), regarding the First Amendment and applying it to the Second. But “the Supreme Court has never applied this presumption [from

Elrod] outside the First Amendment context.” Nat’l Ass’n for Gun Rights v. Lamont, 153 F.4th 213, 248 (2d Cir. 2025).

• Plaintiffs cited Oppenheimer Fund, Inc. v. Sanders, 437 U.S.

340,351 (1978), for the proposition that they are entitled to limited discovery on jurisdictional issues under Rule 26. Oppenheimer, however, was superseded by statute. It is therefore “inappropriate to continue to cite to Oppenheimer for the purpose of construing the scope of discovery under amended Rule 26(b)(1).” Peterson v. Corby, 347 F.R.D. 192, 195 (E.D. Mich. 2024). Any reasonable inquiry – even a very brief one – would have revealed that reliance on Oppenheimer was misplaced.

As this court has previously pointed out, the use of AI tools without independent fact-checking and citation-checking unquestionably violates Rule 11. Chatbots do not conduct a “reasonable inquiry” into existing law or the evidentiary support for factual contentions. A large language model (LLM) is a “word guesser” – it is “trying to predict what the next words would be if that sentence appeared on the internet.” Seither & Cherry Quad Cities, Inc. v. Oakland Automation, LLC, No. 23-11310, 2025 LX 205874, 2025 WL 2105286, at *5 (E.D. Mich. July 28, 2025). It has “no way of ‘knowing’ whether that sentence it created about the case’s holding was in fact true or false.” Id.

In light of the requirements of Rule 11 and the Magistrate Judge’s prior warning (ECF No. 168, PageID.3183), the court ordered Plaintiffs

to show cause in writing why they should not be sanctioned pursuant to Rule 11(b)(2) & (c) of the Federal Rules of Civil Procedure, for attributing fictitious holdings to court decisions, apparently in reliance on AI-generated content, and after clear notice by the court. See, e.g., Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. June 22, 2023); ECF No. 168, PageID.3182 (warning to Plaintiffs).

In response, Plaintiffs, to their credit, generally do not try to avoid fault and admit that they used Chat GPT “without proper verification.” ECF No. 174, PageID.3229. However, two specific points are worth addressing.

First, the fact that Plaintiffs, and particularly Michael Jarrus (who drafted the briefing at issue), did not “fabricate cases or cite nonexistent decisions” is of no help. When a case cite is “real,” an attorney, or for that matter a judge, might see a case they recognize and assume the quote or holding has been accurately represented. Seither & Cherry, 2025 WL 2105286, at *1. That problem is illustrated here; although Chat GPT generated “holdings” that looked like they could plausibly have appeared in the cited cases, in fact it overstated their holdings to a significant degree. And while a litigant might get away with similar overstatements because they could, perhaps, reason their

way to showing how a case’s stated holding might extend to novel situations, an LLM does not reason in the way a litigant must. To put it in a slightly different way, LLMs do not perform the metacognitive processes that are necessary to comply with Rule 11. LLMs are tools that “emulate the communicative function of language, not the separate and distinct cognitive process of thinking and reasoning.” Benjamin Riley, Large language mistake, The Verge https://www.theverge.com/ai-artificial-intelligence/827820/large-language-models-ai-intelligence-neuroscience-problems [https://perma.cc/7EHD-PLLZ]. When an LLM overstates a holding of a case, it is not because it made a mistake when logically working through how that case might represent a “nonfrivolous argument for extending, modifying, or reversing existing law or for establishing new law;” it is just piecing together a plausible-looking sentence – one whose content may or may not be true.

Second, the court is also concerned by, and therefore highlights, the conduct of Plaintiff Linda Jarrous. Plaintiff Michael Jarrus admits to being the primary drafter of the problematic submissions, and the person who originally created and copied the AI-generated text without verifying the truth of its content. But Rule 11 imposes obligations on each and every signatory to a briefing, not just the person who was the

primary drafter. See Seither & Cherry, 2025 WL 2105286. Linda Jarrous submitted her declaration and stated that she relies on her son, Michael Jarrus, to draft the documents, which she reviews “only at a very general level” before she signs. ECF No. 173, PageID.3222. This is insufficient to meet the requirements of Rule 11, and Plaintiff Linda Jarrous is warned that the Federal Rules of Civil Procedure impose obligations on her that are independent of whatever drafting her son does for their joint case; Michael Jarrus is not an attorney and cannot represent her or give legal advice. She will be held equally responsible for the use of misrepresented or otherwise improper citations. A second instance of Plaintiff Linda Jarrus’ failure to independently verify the claims, defenses, factual contentions, or other legal contentions presented in a briefing with her signature may result in dismissal of her claims in their entirety. For now, the court begins with less-restrictive alternatives.

The court is cognizant that imposing a monetary sanction on plaintiffs who qualify for IFP status may be ineffective. See Haskins v. Dart, 633 F.3d 541, 544 (7th Cir. 2011); Simmons v. Gilmore, No. 2:17-CV-00996, 2020 WL 4016331, at *4 (W.D. Pa. July 16, 2020). However, the court also ordered Plaintiffs to explain how much, per month, they

spend on “AI” subscriptions per month. Plaintiff Michael Jarrus

explained that he pays approximately $20 per month in a subscription

to “ChatGPT Plus.” Over the course of 12 months, the evidence

suggests that Plaintiff Michael Jarrus is at least able to afford a Chat

GPT subscription of about $240. Absent proof that a monetary sanction

will prove impossible to pay, the court will enter sanctions sufficient to

deter similar conduct in the future.

It is therefore ORDERED that:

1) Plaintiffs’ objections (ECF No. 170) are OVERRULED for

failure to comply with the court’s order and for violation of Rule 11. See Fed. R. Civ. P. 11 advisory committee’s note to 1993 amendment (“The court has available a variety of possible sanctions to impose for violations, such as striking the offending paper . . .”). In order to leave a public record of Plaintiffs’ noncompliance, the court will not strike the objections, but consistent with the appropriate sanction, does not consider them further on their merits.

2) Consistent with Magistrate Judge Patti’s warning that each AI

citation might incur a cost of $200 per citation, the court adopts that amount and imposes a fine of $300 per Plaintiff (a total of $600) for three misrepresented, AI-generated citations. Each Plaintiff shall, individually, be responsible for paying $300. These fines are due to the Clerk of Court and shall be paid in full by February 2, 2026. Failure to pay these amounts may result in dismissal of this action in its entirety or, if one Plaintiff pays their fine but not the other, of the nonpaying Plaintiff’s claims for failure to comply.

3) If Plaintiffs file any future briefing in this case with even a

single misrepresented, misquoted, or fictitious case that is caused by the use of generative “AI”, this court will strongly consider any recommendation to dismiss this case for bad faith failure to comply with court orders, or revocation of Plaintiffs’ IFP status,1 or in the alternative, it would not be clearly erroneous for the Magistrate Judge to strike or otherwise refuse to consider the merits of an entire briefing for the inclusion of a misrepresented, misquoted, or fictitious case.

SO ORDERED.

Date: December 2, 2025 s/F. Kay Behm

F. Kay Behm United States District Judge