Florida Attorney General James Uthmeier has asked a state court for an emergency order that would sharply restrict OpenAI’s work in the state. The requested injunction would prevent the company from training new artificial-intelligence models without independent safety oversight and would cut minors off from ChatGPT.

The motion is the latest escalation in Florida’s action against OpenAI. Uthmeier’s office began a criminal investigation in April and sued the company in June after the 2025 mass shooting at Florida State University. The suspect allegedly used ChatGPT before the shooting. That allegation is central to the state’s broader argument that the company’s safeguards, product presentation and data practices require court intervention.

OpenAI has previously disputed that its chatbot bears responsibility for the crime. In a statement issued after Florida opened its investigation, the company described ChatGPT as a general-purpose service used by hundreds of millions of people for legitimate purposes. It said it continually works to improve protections intended to identify harmful intent, limit misuse and respond when safety risks emerge.

What Florida is asking the court to do

An injunction is a court order directing a party to do, or stop doing, a particular thing while a case proceeds. It is not the same as a final ruling on whether Florida’s underlying allegations are proved. Here, Uthmeier is seeking immediate restrictions rather than waiting for the lawsuit’s ultimate resolution.

The requested order, as described by the attorney general, has several major elements:

  • No new OpenAI model development without independent safety guardrails or oversight.
  • No access to ChatGPT for minors.
  • No collection of children’s data.
  • No characterization of the product as safe, accurate or reliable.
  • No presentation of the product as human.
  • No engagement tactics designed to keep people conversing beyond a point of danger.

Those asks combine several distinct policy disputes under one emergency filing. Some concern the development process for future systems; others concern how an existing consumer product is marketed and used. Others focus on minors and data collection. The result is a request that reaches beyond a single allegedly dangerous conversation and into how OpenAI builds, describes and operates its AI services.

In a video shared on X, Uthmeier framed the motion as a demand for independent safety measures and family protection. He also challenged OpenAI chief executive Sam Altman to support the court request if he is serious about slowing frontier-model development.

Why the request focuses on “frontier” AI

The term frontier model generally refers to highly capable, leading-edge AI systems whose potential uses and risks may be broader than those of smaller or more narrowly designed tools. In this dispute, Florida’s position is not simply that a chatbot response can be wrong or harmful. The state is arguing that development of increasingly capable systems should not continue without external safeguards.

Related coverage includes Florida Seeks Court Order to Halt New OpenAI Model Training.

That is a substantially more consequential proposition than asking a company to revise one feature. A court-ordered pause on training new models, if granted in the form requested, could affect the cadence of research and product development. The source material does not establish what exact oversight structure Florida wants, who would serve as an independent reviewer, or what standard OpenAI would need to meet before development could resume. Those details would matter greatly in any practical enforcement of the request.

Florida’s brief also invokes OpenAI’s public security disclosures, including one made after the hacking of Hugging Face, along with Altman’s calls for the AI sector to slow frontier development. The state argues that OpenAI’s own public caution supports intervention. Its filing portrays the injunction as consistent with warnings the company and its leadership have themselves made about advanced AI risks.

That remains Florida’s legal theory, not a judicial finding. The court will need to assess the evidence and the legal standards for emergency relief before deciding whether any order is warranted.

Safety claims, product claims and the limits of a general-purpose tool

OpenAI’s prior response highlights an important distinction in the case: a general-purpose AI system can be used for a vast range of ordinary activities, while critics may contend it needs stronger systems to detect and interrupt harmful uses. The company says it already works to detect dangerous intent and limit misuse. Florida argues those efforts are inadequate and wants an outside check on safety before further model development.

Both positions can coexist as descriptions of the dispute without resolving its central question. A tool may have legitimate widespread uses, and policymakers may still examine whether it has sufficient protections in high-risk circumstances. The difficult legal and technical questions concern what safeguards are feasible, how reliable they must be, how they should work for young users, and whether a state court can impose the limits Florida proposes.

The attorney general’s request to stop OpenAI from describing ChatGPT as “safe, accurate, or reliable” also goes to the language surrounding AI. Generative AI systems produce responses by identifying patterns in data and generating likely continuations; they are not inherently guaranteed to provide correct information. This does not, by itself, decide whether any particular product claim is unlawful. It does explain why claims about accuracy and reliability have become a focal point in AI policy arguments.

Similarly, the request to stop the company from “pretending” the service is human is about anthropomorphic design and communication. Conversational software can sound personal even though it is not a person. Florida’s motion treats that distinction as relevant to safety, especially when users may be emotionally vulnerable or may not fully appreciate a system’s limitations.

Minors and data are a separate front in the proposed order

The demand to exclude minors from ChatGPT is among the clearest proposed restrictions, but the supplied record does not lay out how Florida would define a minor-access ban or how it believes such a restriction should be verified. It also does not specify the data practices that Florida alleges violate the law. Still, the filing links children’s access and data collection to its larger claim that current protections do not adequately protect Florida families.

For parents, educators and younger users, the immediate practical takeaway is not that access has changed: Florida has requested an order, and the supplied information does not say one has been granted. The filing instead signals that age gating, child-data handling and safety controls may become central issues in the litigation.

More broadly, the case illustrates why AI regulation frequently bundles together several issues that are easy to blur: privacy, consumer protection, technical safety, platform design and criminal misuse. Each has different factual and legal tests. A data-practice claim is not the same as a claim about model training; a claim about a harmful user interaction is not automatically a finding that an AI provider caused a crime. The court process will have to separate those questions even if the emergency motion presents them together.

A wider debate over warnings and reporting

Florida’s lawsuit is not the only matter linking OpenAI to concerns over violence and chatbot safety. In February, Canadian regulators called on the company to change how it approaches threats of harm after reporting alleged that OpenAI flagged the account of the 2025 Tumbler Ridge shooter Jesse Van Rootselaar but did not alert law enforcement. Following that tragedy, OpenAI agreed to new policies for working with Canadian law enforcement.

That episode points to another complicated issue: what a company should do when its systems detect a possible threat. Detection alone does not answer when a platform can identify a user, how confidently it can assess intent, what information it holds, or when it should contact authorities. Yet the stakes are unusually high when a potential threat of violence is involved. Florida’s filing places that broader safety debate in a state-court fight over model development and ChatGPT access.

The case also arrives as AI companies face scrutiny over how they communicate risk. The same pressure is visible across the technology sector, where advanced systems are increasingly treated not only as consumer software but as infrastructure requiring governance. For related coverage of the business push around AI services, see Meta’s enterprise AI expansion.

What happens next

At this stage, the significant development is the request itself. Florida is asking a court to impose extraordinary operational restrictions on one of the most prominent AI companies, based on its allegations about safety failures, children’s data, product descriptions and risk management. OpenAI did not immediately respond to a request for comment on the new motion.

Nothing in the filing described here means that a court has accepted Florida’s claims, that ChatGPT access has been cut off for minors, or that OpenAI has been ordered to stop training models. Those are requested remedies. The eventual decision on emergency relief will determine whether the dispute produces immediate restrictions or continues through the ordinary course of the underlying lawsuit.

For now, the motion puts a stark question before the court: whether public concerns about advanced AI safety justify an immediate, independently supervised brake on a company’s future model development. Florida says yes. OpenAI’s prior position is that its product is not responsible for the Florida State University shooting and that it already works continuously to strengthen protections against harmful misuse.