
By Jeffrey T. Donner, Esq.
September 30, 2026
When Florida filed its lawsuit against OpenAI in June, I wrote that the case reflected a larger jurisprudential problem: the increasing tendency of government lawyers to ask courts to accomplish through litigation what legislatures have not enacted through law.
The September 30 filings by OpenAI have sharpened that concern. What is striking is not simply that OpenAI disputes the Attorney General’s claims. That was inevitable. More significant is that OpenAI’s counsel has now advanced substantially the same institutional objection I identified when the case was filed: public nuisance cannot become an all-purpose vehicle for regulating lawful products and technologies, and the judiciary’s equitable powers should not be transformed into a substitute lawmaking apparatus merely because a new technology presents difficult social questions.
That is the real issue. The controversy is not simply about ChatGPT, children, artificial intelligence, or the propriety of particular safeguards. It concerns the character of the common law itself, the historical boundaries of public nuisance, and the constitutional distinction between adjudicating legal rights and making prospective regulatory policy.
Public Nuisance Had a Legal Meaning Before It Became a Litigation Strategy
Public nuisance is ancient, but antiquity should not be confused with elasticity without limit. Historically, nuisance law addressed interference with rights held in common by the public and, in many of its most familiar applications, interference associated with land, roads, waterways, pollution, smoke, odors, noise, disorderly premises, gambling establishments, prostitution, obstruction of public ways, and similar conditions affecting the community as a whole.
Florida law still bears that imprint. Section 823.05, Florida Statutes, identifies various buildings, houses, places, and locations maintained for specified unlawful uses as nuisances. Section 60.05 provides an equitable mechanism for abating nuisances associated with buildings or grounds. And the Legislature’s definition of nuisance in the Right to Farm Act expressly concerns interference with the reasonable use and enjoyment of land, including noise, smoke, odors, dust, fumes, particle emissions, and vibration. §§ 60.05, 823.05, 823.14, Fla. Stat. (2026).
That history does not mean that nuisance law was always limited to the paradigmatic case of smoke drifting across a property line. Public nuisance was broader than private nuisance and historically encompassed interference with public highways, waterways, public health, and public morals. But the doctrine nevertheless possessed an intelligible legal core. It was not simply a synonym for conduct that government officials believed was harmful, socially undesirable, or insufficiently regulated.
That distinction is critical because the modern temptation is to convert public nuisance into a residual cause of action that exists whenever more conventional law does not produce the desired regulatory result. If a lawful product is alleged to impose widespread social costs, nuisance is invoked. If a communications platform allegedly contributes to psychological harms, nuisance is invoked. If a new technology creates risks that have not yet been addressed comprehensively by statute, nuisance is invoked again.
At that point, the doctrine ceases to perform a recognizably common-law function. It begins to operate as an open-ended delegation of regulatory authority to courts.
Traditional causes of action impose discipline precisely because they have elements. Negligence requires duty, breach, causation, and legally cognizable injury. Fraud requires a false statement or actionable omission, the requisite state of mind, reliance, causation, and damages. Product-liability doctrines impose their own substantive requirements. Statutory claims depend upon the statute the Legislature actually enacted. An unconstrained nuisance theory appears attractive because it can seem to bypass those constraints: identify a serious social problem, characterize it as harmful to the public, label it a nuisance, and ask a court to abate it.
But if that becomes the governing principle, the judiciary has acquired something close to a general regulatory jurisdiction.
Holmes and Cardozo Understood Development Without Judicial Legislation
The tension between legal development and judicial restraint would have been familiar to both Oliver Wendell Holmes Jr. and Benjamin Cardozo. Neither believed that the common law was static, and neither conceived of judging as a purely mechanical exercise.
Holmes famously wrote in The Common Law that “[t]he life of the law has not been logic: it has been experience.” The phrase is often quoted as though Holmes were endorsing unstructured judicial pragmatism. Read in context, however, Holmes’s point was historical. Legal rules emerge from accumulated experience, inherited doctrine, institutional practice, and social change. To understand what a legal doctrine is becoming, one must understand what it has been.
That is development, but it is development with ancestry.
Cardozo’s The Nature of the Judicial Process made the institutional point more explicitly. Cardozo rejected the fiction that judges simply discover preexisting answers through formal deduction. He recognized that hard cases require judgment, that precedent sometimes runs out, and that courts inevitably participate in the development of legal principles. Yet his theory of judging was not a theory of free-form policymaking. Judicial development remained constrained by precedent, analogy, history, custom, logical coherence, social experience, and the institutional position of the court.
The legitimacy of the common-law method lies in precisely that discipline. Courts extend doctrine incrementally, through concrete disputes, by identifying relevant analogies and testing new applications against inherited principles. They do not ordinarily begin with a desired regulatory result and then stretch an old cause of action until it can support that result.
That distinction becomes especially important when courts confront new technologies. Novel facts do not eliminate the need to identify the governing legal category. If anything, novelty makes doctrinal discipline more important because the pressure to solve a perceived social problem can make legal categories seem inconveniently narrow.
The first question should remain: what law actually governs this conduct? The question should not become: what old doctrine can be expanded sufficiently to produce the policy outcome we now prefer?
ChatGPT Is Not a Traditional Public Nuisance
That problem is evident in the Florida Attorney General’s lawsuit against OpenAI.
The State alleges serious harms associated with ChatGPT and artificial intelligence. Those allegations should be evaluated according to the legal standards governing the causes of action actually pleaded. But the public-nuisance count presents a threshold conceptual difficulty.
There is no obstructed public road. There is no contaminated river or waterway. There is no parcel emitting smoke, sewage, dust, fumes, or noise. There is no disorderly house, gambling hall, brothel, or drug premises being maintained in a neighborhood. There is no physical place whose operation interferes with a right common to the surrounding public.
Instead, the alleged harms arise principally through interactions between individual users and software.
That does not make the alleged harms insignificant. It does, however, raise the prior legal question whether public nuisance is the correct doctrine through which to address them.
OpenAI’s September 30 motion to dismiss makes essentially that argument. It contends that Florida nuisance law has historically concerned land, property, public ways, natural resources, and particular premises; that alleged injuries suffered by individual ChatGPT users do not become an interference with a public right merely because many people allegedly experience them; and that individualized injuries ordinarily should be addressed through causes of action belonging to those who actually suffered them.
The point is not novel. In Penelas v. Arms Technology, Inc., 778 So. 2d 1042 (Fla. 3d DCA 2001), Miami-Dade County attempted to use public nuisance, negligence, and related theories to obtain relief that would effectively regulate firearm design and distribution. The Third District rejected that effort and warned that the judiciary may not “‘enact’ regulatory measures in the guise of injunctive relief.” Id. at 1045.
Penelas arose in a different statutory setting, including an express legislative preemption provision governing firearms. It therefore should not be overstated. But its institutional reasoning remains powerful because it addresses the precise danger presented whenever litigation shifts from enforcing existing law to prescribing the prospective rules by which a lawful industry must operate.
Florida Supreme Court nuisance law reflects a similar insistence on limits. In Keshbro, Inc. v. City of Miami, 801 So. 2d 864, 876 (Fla. 2001), the Court emphasized that nuisance abatement must be tailored to eliminate objectionable conduct without unnecessarily infringing upon lawful enterprise. The point is not that Keshbro decides an artificial-intelligence dispute. It does not. The significance lies in the Court’s understanding that nuisance abatement is remedial and bounded, not an invitation to restructure lawful activity more broadly than the nuisance itself requires.
Modern appellate courts outside Florida have confronted related efforts to use nuisance law as a generalized response to products associated with widespread social harm. In State ex rel. Hunter v. Johnson & Johnson, 499 P.3d 719 (Okla. 2021), the Oklahoma Supreme Court rejected an opioid public-nuisance judgment and emphasized the distinction between communal injury and an aggregation of individual harms. Likewise, in State v. Lead Industries Ass’n, Inc., 951 A.2d 428 (R.I. 2008), the Rhode Island Supreme Court rejected an expansive public-nuisance theory directed at lead-paint manufacturers. These decisions are not binding in Florida, and the jurisdictions are not uniform. California, for example, permitted a broader public-nuisance theory in People v. ConAgra Grocery Products Co., 17 Cal. App. 5th 51 (2017). But the disagreement itself illustrates the jurisprudential question: whether nuisance law remains tied to interference with public rights or becomes a general mechanism for regulating lawful products that allegedly create widespread harms.
The Separation of Powers Is a Practical Doctrine
The broader constitutional framework reinforces the same concern.
Article II, section 3 of the Florida Constitution divides governmental power among the legislative, executive, and judicial branches. That principle is sometimes treated as an abstract structural rule. In practice, it determines who has authority to make fundamental policy choices.
In Askew v. Cross Key Waterways, 372 So. 2d 913 (Fla. 1978), the Florida Supreme Court emphasized that “fundamental and primary policy decisions” must be made by the legislators elected to make them. Id. at 925. Askew involved legislative delegation rather than nuisance law, so it does not mechanically resolve this controversy. But its institutional premise is directly relevant: foundational policy choices belong to the lawmaking branch.
The Court reiterated that principle in Chiles v. Children A, B, C, D, E & F, 589 So. 2d 260 (Fla. 1991), explaining that one branch may not exercise powers constitutionally assigned to another. Again, the factual context was different. The significance is structural. A judicial proceeding does not become a legislative process merely because the subject is important and the requested relief is prospective.
Artificial intelligence demonstrates why that distinction matters.
Should every person under eighteen be prohibited from using generative AI? Should parents be permitted to authorize supervised use by older minors? Should a seventeen-year-old be barred from using AI to study calculus, programming, history, or a foreign language? Should different age bands receive different treatment? Should frontier models require independent approval before development? Who chooses the reviewing body? What standards would it apply? What constitutes an adequate safety “guardrail”? May an AI system use first-person language? What warning must appear before use? When does a clarifying question become impermissible “conversation prolongation”? Should professional users be governed by the same restrictions as children or casual consumers?
Those are not merely factual questions waiting to be adjudicated under an established legal standard. They are regulatory choices requiring line drawing, empirical study, expert testimony, definitions, exceptions, enforcement mechanisms, administrative structures, and continuing revision as the technology changes.
A legislature is institutionally designed to perform that function. A trial court ruling on a motion for temporary injunction is not.
The Difference Between Applying Law and Supplying Law
None of this places artificial-intelligence companies beyond existing law. If an AI company commits actionable fraud, courts can adjudicate fraud. If it violates FDUTPA, courts can enforce FDUTPA. If an individual plaintiff proves duty, breach, causation, and injury, courts can adjudicate negligence. If the Legislature enacts privacy protections for children, courts can enforce those protections. Novel technology does not suspend ordinary law.
The distinction is between applying law and supplying law.
That distinction becomes blurred when a broad equitable doctrine is used to fill perceived gaps in a statutory regime. If no legislature has determined that AI development must be conditioned on independent third-party approval, a court should be cautious about creating such a regime through a preliminary injunction. If lawmakers have not categorically prohibited all minors from using generative AI, the question whether every seventeen-year-old in Florida should be barred from doing so entails policy choices that extend well beyond adjudication of a discrete deceptive act.
The same distinction applies to compelled product language. Prohibiting materially false statements is conventional adjudication. Requiring a company affirmatively to characterize its product as “unsafe, unreliable, and inaccurate” is qualitatively different. Likewise, judicial regulation of whether an AI system may use the first person or ask follow-up questions reaches into the architecture and conversational design of the product itself.
The broader the remedy, the more acute the institutional concern.
Public Nuisance as an Ersatz Regulatory Code
The modern appeal of public nuisance is therefore understandable. It promises breadth where traditional causes of action impose doctrinal limits.
For government plaintiffs, that flexibility can be tempting. If negligence requires a particular plaintiff to establish duty, breach, causation, and injury, nuisance appears to avoid some of those limitations. If statutory causes of action are too narrow, nuisance may seem capable of reaching further. If lawmakers have not enacted a desired rule, equitable abatement may appear to offer a judicial substitute.
But that is precisely why the doctrine requires boundaries.
A legal rule framed as “conduct producing widespread negative externalities” is not really a tort element. It is a policy category. Almost every heavily regulated product or industry could be placed within it depending upon how broadly costs are defined. Automobiles cause death and pollution. Alcohol causes addiction and public-health costs. Social-media platforms may affect psychological health. Pharmaceuticals create serious risks alongside extraordinary benefits. Financial products can contribute to systemic instability. Artificial intelligence likewise presents both extraordinary utility and genuine risks.
The mere existence of social costs does not tell us which conduct should be unlawful, what safeguards should be required, or which tradeoffs society should accept. Those are policy judgments.
That is why OpenAI’s September 30 motion is significant when it argues that, if conventional causes of action do not adequately address emerging technological harms, the appropriate institutional response is legislation capable of accounting for all affected interests rather than indefinite expansion of nuisance law. The motion expressly invokes the Restatement (Third) of Torts for that proposition and then argues that regulation of lawful AI technologies belongs to the Legislature rather than being constructed through nuisance abatement.
That is substantially the concern I identified when the Florida lawsuit was filed.
I found the convergence reassuring. My June article was not written in response to OpenAI’s defense papers; those papers did not yet exist. I approached the issue from the perspective of a Florida lawyer who has spent his career working with common-law causes of action and who also uses generative artificial intelligence extensively in legal practice. The institutional problem seemed evident then, and OpenAI’s later briefing confirms that experienced counsel examining the same lawsuit independently identified the same tension between adjudication and regulation.
Holmes and Cardozo Were Not Enemies of Legal Change
Defending doctrinal limits is sometimes caricatured as hostility to legal development. Holmes and Cardozo demonstrate why that is mistaken.
Neither believed nineteenth-century rules should be mechanically frozen and applied forever to facts the nineteenth century could not have imagined. Both recognized that law changes because society changes. Yet the legitimacy of common-law development depends on how that change occurs: concretely, incrementally, analogically, and with continuing regard for inherited doctrine.
There is a profound difference between extending an established legal principle to a genuinely analogous new circumstance and converting an old doctrine into an unrestricted judicial power to solve contemporary social problems. The former is common-law reasoning. The latter is legislation by another name.
The distinction cannot be eliminated simply by describing both processes as “development of the common law.” Cardozo’s recognition that judges sometimes make law at the margins was not a theory that judges possess general legislative competence. Holmes’s emphasis on experience was not an argument that history ceases to matter whenever a new policy problem emerges.
The common law survives precisely because it is adaptable without becoming formless.
Artificial Intelligence Requires Legislative Judgment
There may ultimately be a substantial need for legislation governing artificial intelligence. The rapid development of increasingly capable models may make comprehensive regulation inevitable.
Many of the most consequential issues are inherently national. AI systems are developed through interstate and international networks, trained and deployed across jurisdictions, and used simultaneously by individuals and businesses across the country and the world. A fragmented system in which fundamentally different design requirements are imposed state by state presents obvious practical and constitutional difficulties. Congress is institutionally positioned to consider national standards concerning frontier-model testing, safety requirements, minors, privacy, autonomous agents, disclosure obligations, cybersecurity, and related issues whose consequences do not stop at state borders.
States nevertheless retain substantial authority in traditional areas such as consumer protection, privacy, tort law, and protection of minors. If Florida concludes that existing law is inadequate, the Florida Legislature can conduct hearings, hear from competing experts, establish age classifications, preserve or limit parental authority, define prohibited practices, create administrative expertise, specify remedies, and revise the governing framework as evidence and technology evolve.
The Legislature can therefore perform the task that legislatures are designed to perform: make the fundamental policy choice, define its scope, establish exceptions, delegate technical implementation where appropriate, and create a framework that can be revised over time. Courts can then perform the task for which they are institutionally suited: interpret the enacted law, resolve concrete disputes, determine whether the legal standard has been satisfied, and grant remedies authorized by that law. That sequence is not a procedural inconvenience to be bypassed when a policy problem appears urgent. It is part of the constitutional division between lawmaking and adjudication.
The alternative is undeniably expedient. A government lawyer can identify a social problem, invoke a sufficiently elastic common-law label, and ask a court to announce the desired regulatory solution through an injunction. That route may be faster than legislation, may avoid political compromise, and may even produce a policy that some judges and lawyers regard as desirable. But expediency is not the measure of constitutional legitimacy, and public nuisance should not become a residual regulatory power available whenever existing causes of action fail to produce the preferred policy outcome.
That is not the common law at its best, and it is not how fundamental public policy should be made. When the problem requires new prospective rules governing lawful conduct, the proper answer is legislation.
