Beginning June 15, 2026, every signature on a Florida court filing will certify that the legal authorities identified in it “exist and are accurately cited.” The Florida Supreme Court added that representation to Rule of General Practice and Judicial Administration 2.515(d)(2) on May 28, 2026, and expressly authorized courts to sanction filings inconsistent with it.
The coverage has treated this as a hallucination rule about fabricated citations, phantom cases, and opinions that don’t exist. That framing is understandable, and incomplete. The harder question, the one the commentary has skipped, is what accurately cited means. And the answer matters most to the party the amendment seems almost designed to overlook, the unrepresented litigant who turned to AI precisely because they couldn’t afford an attorney.
Start with what the amendment does not do. It does not require disclosure of AI use. It does not ban AI from legal practice. It does not mandate any particular verification workflow. It asks the signer to certify that the cited authorities exist, and that they are accurately cited. That narrowness is a choice, and a defensible one. The circuit-level patchwork that preceded the amendment (disclosure requirements of varying scope across Miami-Dade, Broward, Palm Beach, and other circuits) regulated the process of using AI. The Supreme Court regulates the output. A lawyer who uses AI responsibly and verifies every citation is treated identically, under this rule, to a lawyer who never touched AI at all. Tool-agnostic is almost certainly the right regulatory design.
But tool-agnostic rules still require interpretation, and “accurately cited” is not self-defining.
The spectrum of inaccuracy
Inaccuracy runs on a spectrum. At one end sits the case that simply does not exist, the fabricated citation that Gouveia v. Meridian Financial Investments, 4th DCA 2026, illustrated with such clarity. That filing fails the exists prong without ambiguity; no reasonable reading of the rule saves it. The spectrum extends well past fabrication, though, and the accurately cited prong is where it gets genuinely complicated.
Consider three scenarios the text has to cover.
In Scenario One, a lawyer cites a real case with a correct citation string, but the parenthetical mischaracterizes the holding. The case held X; the parenthetical says it held Y. A minimal reading of the rule, focused on citation format and existence, might call that accurate. The broader reading, that a citation accurately represents what the authority says, almost certainly does not.
In Scenario Two, a lawyer cites a real case that arguably supports the cited proposition, but the case is distinguishable on material facts, and a careful reader would dispute whether it stands for what the brief claims. This is the hardest case and the most common one. It is advocacy, and it shades into mischaracterization depending on the degree of the stretch.
In Scenario Three, a lawyer cites a real case with a correct citation string, describes the holding accurately, but the case has been distinguished into near-irrelevance by subsequent opinions the brief does not disclose. The citation is accurate; the picture it paints is not. That is less an “accurately cited” problem than a candor problem under Rule 4-3.3(a)(3). The line between the two is not obvious.
These are not hypothetical edge cases. They are the daily texture of appellate brief writing, and AI makes them more likely, not because AI fabricates, but because AI summarizes, and summaries elide qualifications. A model that reads a case and concludes it stands for a broad proposition is not hallucinating. It is wrong about the scope of a holding, the way a first-year associate is sometimes wrong.
The amendment’s text does not say which scenarios violate the certification. The per curiam opinion says only that the amendment is aimed at “the demonstrated risks of generative AI” and the need to “promote the accuracy and integrity of court filings,” which suggests the drafters were thinking primarily about fabrication. But the words they chose reach further. Accurately cited comfortably covers Scenario One and plausibly covers Scenario Two. So until the Court says otherwise, treat the certification as meaning what it says in full. It means the citation exists, the case stands for the proposition cited, and the characterization in the parenthetical is fair. If you cannot verify that from the case itself, the parenthetical does not belong in the brief.
The pro se problem
The amendment extends to unrepresented parties, and that extension deserves more scrutiny than it has received.
The institutional logic is easy to state. Courts have an interest in accurate citations regardless of who submits the filing; a fabricated citation from a pro se litigant wastes judicial resources just as one from a licensed attorney does; and a certification with teeth only against lawyers creates an asymmetry that sophisticated bad actors could exploit. Grant all of that. The practice is harder.
AI tools were supposed to democratize access to legal process. The pitch (explicit in the marketing of every major legal AI platform, implicit in the spread of consumer tools like ChatGPT and Gemini) is that unrepresented individuals can now research legal questions, draft pleadings, and navigate courts without retaining counsel. That pitch has always had an asterisk, but the asterisk was understood to be the quality of the output may vary, not if the output is wrong you may lose your case.
Under the amended rule, a pro se litigant who uses a free AI tool, receives a citation to a real case that is inaccurately described, and incorporates it into a motion without the legal training to recognize the error now faces sanctions that include dismissal of the action. The available penalties (reprimand, contempt, striking the filing, dismissal, fees and costs, or “other sanctions”) are calibrated for attorneys who knowingly or recklessly submitted false authority. They do not obviously fit a litigant who genuinely did not know their AI had gotten a case wrong.
Courts retain discretion, and nothing in the amendment requires dismissal in every case of inaccurate citation. But the rule as written is strict in its terms. A filing inconsistent with the certification is sanctionable; whether the signer knew the citation was inaccurate is not an element of the violation as the text describes it. It draws no line between represented and unrepresented parties, and it says nothing about intent.
The deeper problem is structural. Attorneys have verification tools (Westlaw, Lexis, Fastcase, FLexlaw) and their own doctrinal training. They know to read the case before citing it. Pro se litigants by definition lack the training, and frequently lack access to the professional-grade research tools that would let them check what an AI told them. The amendment assumes a capacity for verification that many unrepresented litigants do not have.
None of this means the rule should exempt pro se litigants. It means courts applying sanctions against unrepresented parties should weigh proportionality and intent with care, and it means the Bar, the Supreme Court, and the public comment process should consider whether clarifying language is warranted.
Before August 11
The comment period closes August 11, 2026, and the Court opened it precisely because it did not circulate the amendment for public feedback before issuance. Two drafting problems sit squarely within its reach.
First, the rule should say whether accurately cited means accurate as to citation format and the case’s existence, or accurate as to the characterization of the holding. The former is workable and clearly scoped. The latter is a meaningful representational obligation (certifying not just that a case exists but that your description of it is correct), and practitioners are entitled to know whether they are signing it.
Second, the rule should address the role of intent in the sanctions analysis for unrepresented parties. A scienter element, even a recklessness standard, would give courts principled guidance and would protect pro se litigants from punishing consequences for honest errors while still deterring knowing misuse of AI tools.
Practitioners who file briefs for a living should weigh in.
Unrepresented parties should not wait for the answer. Treat every AI-generated citation as unverified until you have confirmed it yourself from a reliable source. Most county clerks of court offer free online docket search, and many county law libraries provide free access to legal research databases. Verification is not always easy without professional tools. An AI-generated brief that costs you your case on a procedural sanction is worse.
The rule is good policy in its essential purpose. The question of what it actually requires, and who it will actually hurt, is still being written.
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. His earlier piece on AI hallucination in Florida courts, analyzing Gouveia v. Meridian Financial Investments (4th DCA 2026), is available in the Research Vault.