Hulse-Gibson v. Hulse, decided July 8, 2026, is two pages long. The Second District Court of Appeal affirmed the orders on appeal “without comment”; the merits, whatever they were, did not detain the panel for a single sentence of analysis. Everything else in the opinion is addressed to one person. “We write, however, to caution Cynthia Hulse-Gibson about using generative AI in any future pro se filings without verifying the accuracy of the results.”
Pause on the shape of that. A district court of appeal published an opinion whose entire substance is a warning, by name, to an unrepresented litigant about her research tool. No sanctions. No show-cause order. No analysis of the case she brought. Two pages, a caution, and a citation to a rule in force for twenty-three days.
That shape is half the story.
Two answers in twenty-three days
In May, after the Florida Supreme Court amended Rule of General Practice and Judicial Administration 2.515(d)(2) to make every signature on a Florida court filing a representation that the authorities cited in it “exist and are accurately cited,” I wrote that the rule was being misread as a hallucination rule, that its hardest interpretive question was what accurately cited actually means, and that the party most exposed was the unrepresented litigant who turned to AI precisely because they couldn’t afford an attorney. The rule, I noted, is strict in its terms (no scienter element, no line between represented and unrepresented parties), and the question of who it would actually hurt was still being written.
It is being written faster than I expected. The amendment took effect June 15, 2026. Eleven days later, the Sixth District Court of Appeal issued Davidson v. Chang-Warner, No. 6D2025-1678 (Fla. 6th DCA June 26, 2026), ordering a pro se appellant to show cause why he should not be sanctioned over an initial brief citing authorities that either do not exist or do not say what they were cited for. Twelve days after that came Hulse-Gibson, twenty-three days after the effective date and thirty-four days before the amendment’s public comment window closes on August 11. Two appellate opinions holding the new certification up to pro se litigants arrived while the ink on the rule is, quite literally, still open for revision.
Two cases under the new words
The precision matters here, because Florida’s appellate courts have been sounding this alarm for a while and the new opinions are easy to misread as more of the same.
They aren’t. In Gleason v. Marcus, 51 Fla. L. Weekly D962 (Fla. 2d DCA May 13, 2026), the court felt compelled to comment on the filer’s apparent use of artificial intelligence and said it was again “sound[ing] the alarm” about fabricated and misdescribed authority, addressed itself to “all litigants—both pro se and represented,” and cited Rule 2.515(d)(2). But Gleason issued May 13, fifteen days before the Supreme Court adopted the amendment and a month before it took effect. The rule Gleason cited contained no citation-accuracy representation at all; the “exist and are accurately cited” language of new subdivision (d)(2)(D) did not yet exist. Gleason invoked the old rule. Davidson and Hulse-Gibson invoke the new one, and they invoke it in strikingly different registers.
The Sixth District went first, and it did not write a caution. Davidson opens flat. “Appellant—a pro se filer—has filed an initial brief that cites non-existent legal authorities or cites legal authorities inaccurately in violation of Florida Rule of General Practice and Judicial Administration 2.515(d)(2).” What follows is not a warning about future filings; it is an order to “show cause why sanctions should not be imposed” within ten days. The order spells out the menu, which is the amended rule’s full sanctions list plus one addition of the court’s own, “a bar on pro se filing in this Court.” (A care point, because this post is about care. Davidson never says the brief was AI-generated. The word appears in the opinion only inside its quotations of Hessert. The failure mode is identical, authorities that do not exist or do not stand for what they are cited for, but the attribution to a machine is not the court’s, and I won’t supply it.)
Hulse-Gibson invokes the same new words to a different end. It quotes the amended text (“[o]n filing, each signer represents that the legal authorities identified exist and are accurately cited”) and points to the range of sanctions standing behind it, and then deploys none of them. The whole opinion runs to caution.
There is a timing question buried in both docket numbers, and it should be handled with the care the courts themselves used. Both appeals are 2025 cases (No. 2D2025-1144 and No. 6D2025-1678), which means the briefing in each very likely predates the rule the opinions quote. The Second District never engages the point; its caution is framed entirely prospectively, about “any future pro se filings,” and the sanctions it describes are potential, not imposed. The Sixth District engaged the point head-on and resolved it against the filer, writing that “[a]lthough Appellant’s initial brief was filed before the amendments to rule 2.515(d)(2) became effective, we see no issue with applying rule 2.515(d)(2) as amended to Appellant’s initial brief.” The amendments, Judge Pratt reasoned across five pages, “merely clarify what already fell within the scope of the pre-existing requirements” (the accuracy duty was always there, the sanctions authority was always there), so the amended rule “is not being applied retrospectively.” And for good measure, trial courts “may also apply rule 2.515(d)(2) as amended to documents that were filed in the trial courts before the amended rule became effective.” Rules, unlike customs, have effective dates. And the first court to confront this one’s effective date held that it is not a shelter. If you signed a filing in 2025 and it is still pending, the new words on the signature line reach back to it, because on the Sixth District’s account they were never new.
So in the amended rule’s first twenty-three days, two district courts saw two materially similar pro se filings under one new rule. One answered with a show-cause order, the other with a caution. Hold that contrast. It is where this post ends up.
Better at being wrong
What earned Ms. Hulse-Gibson the caution was not the classic hallucination docket of 2023, invented cases with invented captions. Her briefs, the court wrote, were “replete with mismatches involving the purported legal proposition, the case name, and the reporter citation.” Real components, wrongly assembled. A genuine proposition was attached to the wrong case, a genuine case to the wrong reporter page. The panel reached out of state for the diagnosis, quoting the Fifth Circuit in Fletcher v. Experian Information Solutions, Inc., 168 F.4th 231, 233 (5th Cir. 2026), which wrote, “This problem [the misuse of generative AI] now often manifests as false quotes or statements of law attributed to real cases, rather than the more easily recognizable fake cases.”
To be fair to the record, this is an emphasis shift, not a new species. Gutierrez v. Gutierrez, 399 So. 3d 1185 (Fla. 3d DCA 2024), the case Hulse-Gibson cites as its sanctions precedent, already featured “phantom case law” in both flavors (fabricated citations alongside real cases quoted for things they never said), and Gleason saw the same mix this spring. What has changed is the ratio, and the ratio is the problem. A fabricated case fails at the first database lookup; anyone with a search box can catch it. A real case miscited for a proposition it does not support fails only when someone reads the case. The machines have not gotten better at being right; they have gotten better at being wrong. And the cost of catching them has migrated from the clerk’s office to chambers. That is precisely the resource drain that turns judicial patience into published warnings.
The court’s own prescription is the sentence pro se readers should tape to the monitor. “Those who use generative AI to assist with legal research and brief writing must exercise special care to validate the results before signing and filing their briefs; their failure to do so exposes them to potential sanctions.”
Who the warning is for
Here is the puzzle. In his concurrence in Gouveia v. Meridian Financial Investments, No. 4D2025-0843 (Fla. 4th DCA Mar. 25, 2026), Judge Lott gave the most clear-eyed account any Florida jurist has offered of why opinions like Hulse-Gibson shouldn’t work. The sanctions toolbox “works well enough for attorneys” (repeat players, professionally trained, with licenses to lose). Pro se litigants are one-shot participants; the next self-represented filer never read the last published warning, and generative AI is, in the line the Second District quoted back in Gleason and that gives this post its title, “very good at sounding right, but less adept at being right.” Lott’s conclusion was that warnings are structurally mismatched to the pro se problem, and that front-end disclosure and certification rules are “probably the right starting point.”
The Second District has read Gouveia (Gleason quotes the concurrence), and it published another warning anyway. It was not its first. Clerk of Ct. & Comptroller for 13th Jud. Cir. v. Rangel, No. 2D2024-1772, 2025 WL 2486314 (Fla. 2d DCA Aug. 29, 2025), and Russell v. Mells, 426 So. 3d 913 (Fla. 2d DCA 2025), handled the attorney version (in Russell, counsel who filed a fabricated case had “fundamentally abdicated” her duty, and was referred to The Florida Bar); Straub v. Henderson, No. 2D2025-1787 (Fla. 2d DCA Feb. 25, 2026), and Gleason handled the pre-amendment pro se version. Hulse-Gibson is at least the court’s fifth entry in under a year, and the same judges keep signing these opinions. Judge Rothstein-Youakim, who wrote Hulse-Gibson, sat on the Gleason panel.
So if Lott is right that the warning will not reach the next pro se litigant, who is it for? Three audiences, I think, none of them hypothetical.
First, Cynthia Hulse-Gibson herself. A warning addressed by name is not rhetoric; it is record. The pro se cases that end badly (Gutierrez ended in dismissal of the consolidated appeals as a sanction and a Spencer bar on future pro se filings in the underlying probate matter) are the ones where the litigant had been told. The named caution converts any repeat performance from an honest error into a knowing one, which is exactly the element the rule’s text omits.
Second, the next court. Trial judges across the Second District now have an appellate opinion, citable in two keystrokes, establishing that proposition-mismatch is sanctionable territory under the amended rule. The warning is a tool left on the bench.
Third (and this is the one that matters this month), the rulemaking file. An opinion issued in the middle of an open comment window, applying the new rule’s language to the exact fact pattern the commentary fought about in the abstract, is evidence whether or not the panel intended it that way. The comment period now has a docket.
The ladder and the gap
Step back and the remedial architecture is visible as a ladder. For pro se filers, courts warn (Goya v. Hayashida, No. 4D2023-1212, 418 So. 3d 652 (Fla. 4th DCA Aug. 13, 2025); Gouveia; Gleason; now Hulse-Gibson), then show cause (Takefman v. Pickleball Club, LLC, 418 So. 3d 826 (Fla. 3d DCA 2025); Hessert v. Hessert, 431 So. 3d 610 (Fla. 6th DCA 2026); now Davidson), then dismiss and bar (Gutierrez). For attorneys, the ladder tops out somewhere else entirely. It ends in Bar referral, as in Russell. That last rung does not exist for the unrepresented (no license, no Rules of Professional Conduct, no disciplinary jurisdiction). This was Lott’s structural point, and it is why Rule 2.515 matters more for pro se litigants than for anyone. The rule binds signers, not members of the Bar. It is the first statewide, rules-based instrument that attaches citation accuracy to the signature itself (“reprimand, contempt, striking of the document, dismissal of proceedings, costs, attorneys’ fees, or other sanctions”), and every unrepresented party signs. The substantive standard was never different; as the Fifth District put it in Stueber v. Gallagher, 812 So. 2d 454, 457 (Fla. 5th DCA 2002), “In Florida, pro se litigants are bound by the same rules that apply to counsel.” What was different was the remedial toolkit. The amendment makes express, statewide, and signature-borne what had been a patchwork of circuit administrative orders and appellate sanctions practice. That is another way of saying the people it newly reaches are the people with no professional apparatus for complying with it.
This is the accountability half of a pairing I wrote about in June. The same season that gave pro se litigants work product protection for their AI preparation is now policing their AI filings. Private use of the tool is yours. The moment its output crosses the signature line, it is the court’s.
The fine print
What these opinions are not, stated plainly. Hulse-Gibson imposed no sanctions; the court’s discretion ran entirely to caution. Davidson has imposed none either. An order to show cause is the beginning of a sanctions inquiry, not its end, and the show-cause process exists precisely so that nothing may come of it. Neither opinion says anything about the underlying dispute (what either litigant sought below, or why they lost), so neither will I. Both should be read as provisional. Hulse-Gibson is a slip opinion, subject to revision before official publication, and Davidson carried the standard not-final-until-rehearing-time-expires notation; that window closed July 11, and I have not seen a rehearing disposition as of publication. Davidson never states that the offending brief was AI-generated (the word appears only in the Hessert passages it quotes), so pairing the two cases as an AI story is partly my framing of the pattern, not a finding either court made. And the observation that Ms. Hulse-Gibson’s briefs likely predate the rule is an inference from the docket year. The Second District never says it. The Sixth District said it outright. Davidson’s brief was filed before the amendment took effect, and the court held that it doesn’t matter.
Twenty-six days
The access-to-justice arithmetic has not changed since May, but it has sharpened. Low-income Americans receive no or insufficient legal help for 92% of their substantial civil legal problems (2022). The pro se share of non-prisoner federal civil filings, steady near 11% for two decades, hit 16.8% in fiscal 2025 (a surge with, in the words of the MIT and USC researchers who measured it, “no precedent in 25 years of administrative records”), and the growth is concentrated by case type, in civil rights complaints, consumer credit disputes, and foreclosure proceedings, the filings where a well-structured factual narrative is most of the work. (My inference, not their finding, is that those are the claims many of these filers could not have brought at all before the tools existed.) The docket where AI-assisted filings concentrate is the docket where lawyers are not. That is not a coincidence; it is the whole phenomenon. And the best available data on how courts actually respond suggests sanctions attach less to the error than to what follows it. They fall on litigants who, in the words of the researcher who maintains the leading tracker of these cases, “refused to own up to it, double-downed, made up stories, or blamed the intern.”
I said in May that none of this argues for exempting pro se litigants from the certification, and neither of these opinions has moved me. A fabricated or miscited authority consumes the same judicial resources whoever signs the filing. But put the two opinions side by side and the thing I did ask for becomes impossible to miss. In the amended rule’s first twenty-three days, two district courts of appeal confronted materially similar pro se filings under it. The Sixth District answered with an order to show cause, which it applied, after five pages of reasoning, to a brief signed before the rule’s effective date. The Second District answered with a published caution and no sanction. Same rule, same month, same class of litigant; the distance between a warning and a sanctions proceeding was covered entirely by which panel the case drew. That is not a criticism of either court. Each response is defensible on its own record. It is a description of the rule. The proportionality lives nowhere in its text. There is still no scienter element, still no line between the knowing and the ignorant, and dismissal is still available for an honest, unrecognized error on the first offense. Everything between caution and catastrophe is judicial discretion. That is precisely the gap I said in May the comment period should address. This month, that gap is not a blog post’s complaint. It is a comment.
The window closes August 11, twenty-six days from today. The Supreme Court opened it precisely because it adopted the amendment without prior publication, and it now has something no commenter had in May. It has actual cases, pointing in different directions. If you file briefs for a living, or build tools for the people who can’t afford someone who does (I do both), Davidson and Hulse-Gibson are the first data points to cite. A named warning, under the new rule, to a litigant who, in her own case, probably signed before the rule existed. And, twelve days earlier, a holding that signing before the rule existed is no shelter at all.
Judge Lott said the technology is very good at sounding right and less adept at being right. So, on its current text, is the rule. There are twenty-six days left to fix that.
The authorities: Hulse-Gibson v. Hulse, No. 2D2025-1144 (Fla. 2d DCA July 8, 2026) (slip op., subject to revision prior to official publication); Davidson v. Chang-Warner, No. 6D2025-1678 (Fla. 6th DCA June 26, 2026) (order to show cause; not final until time expires to file motion for rehearing); Gleason v. Marcus, 51 Fla. L. Weekly D962 (Fla. 2d DCA May 13, 2026); Gouveia v. Meridian Financial Investments, LLC, No. 4D2025-0843 (Fla. 4th DCA Mar. 25, 2026) (Lott, J., concurring); Gutierrez v. Gutierrez, 399 So. 3d 1185 (Fla. 3d DCA 2024); Russell v. Mells, 426 So. 3d 913 (Fla. 2d DCA 2025); Goya v. Hayashida, No. 4D2023-1212, 418 So. 3d 652 (Fla. 4th DCA Aug. 13, 2025); Stueber v. Gallagher, 812 So. 2d 454, 457 (Fla. 5th DCA 2002); State v. Spencer, 751 So. 2d 47 (Fla. 1999); Fletcher v. Experian Information Solutions, Inc., 168 F.4th 231, 233 (5th Cir. 2026) (reporter citation as given in the Hulse-Gibson slip opinion); Fla. R. Gen. Prac. & Jud. Admin. 2.515(d)(2), as amended by In re Amendments to Florida Rule of General Practice and Judicial Administration 2.515, No. SC2026-0673 (Fla. May 28, 2026) (effective June 15, 2026; comments due August 11, 2026). Quotations were verified against the slip opinions and rule text as filed; the justice-gap figure is the Legal Services Corporation’s, and the pro se filing statistics are from Shah & Levy, Access to Justice in the Age of AI (working paper, Mar. 2026). A fuller chronology of Florida’s AI-citation cases is in this survey.
John J. Lloyd is a Florida/Texas litigator and legal technologist, and the creator of FLexlaw, a Florida-focused legal research platform. He writes about Florida litigation and legal technology at lloydesq.com.