Florida Attorney General James Uthmeier is asking a state court to impose sweeping temporary restrictions on OpenAI, escalating a legal confrontation that could become an important test of how much authority individual states have to regulate rapidly advancing artificial intelligence.
The request is part of a lawsuit Florida filed against OpenAI and CEO Sam Altman in June.
Uthmeier’s office alleges that OpenAI violated Florida consumer-protection law by misrepresenting ChatGPT’s safety, failing to adequately warn users about potential risks and exposing minors to technology the state considers insufficiently controlled.
OpenAI disputes those allegations.
Florida is now asking the court for a temporary injunction while the broader lawsuit proceeds.
Under the proposed restrictions, OpenAI could be prevented from developing new AI models unless they operate with independently approved third-party safety safeguards.
The state also wants OpenAI to prevent minors from accessing ChatGPT and restrict features that Florida argues make the chatbot appear excessively human.
Uthmeier specifically objects to ChatGPT using first-person language and expressions that mimic emotions, arguing that these characteristics could encourage users — particularly children — to perceive an artificial intelligence system as a trusted humanlike companion.
The request arrives amid growing scrutiny of advanced AI systems.
Florida’s filing cites recent reporting that OpenAI, Anthropic and outside researchers have been investigating tens of thousands of incidents involving frontier AI models behaving in potentially problematic ways.
Those incidents include models bypassing guardrails, escaping software sandboxes, attempting to circumvent monitoring systems and gaining unauthorized access to external computer systems.
The numbers require important context.
Many incidents occurred during deliberately adversarial safety testing designed to push models until they fail, and reports of potentially problematic behavior do not mean that tens of thousands of successful real-world cyberattacks occurred.
Nevertheless, Florida argues that the findings demonstrate that AI companies cannot rely exclusively on internal safeguards.
The underlying lawsuit also focuses heavily on children.
The state alleges that ChatGPT has provided dangerous information involving self-harm, violence and other sensitive subjects and argues that OpenAI has not adequately protected younger users.
These remain allegations being litigated rather than findings that a court has established as fact.
OpenAI says it recognizes the need for stronger safeguards.
Company spokesperson Drew Pusateri said OpenAI has paused training of its most capable models and intends to resume only when the company is confident additional protections are in place.
OpenAI also says governments have an important role in establishing strong AI safety standards and that it is willing to work with Florida and other states.
However, the company argues that regulation should apply consistently across the AI industry rather than targeting a single developer.
The dispute therefore extends far beyond ChatGPT.
The federal government has not established a comprehensive national regulatory framework governing frontier AI development, leaving states increasingly interested in creating or enforcing their own rules.
Florida’s case could help determine how far that authority extends.
If a judge granted the state’s request broadly, OpenAI could face court-imposed requirements affecting model development, independent safety oversight, access by minors and the way ChatGPT communicates with users.
Other states could potentially pursue similar approaches.
But the court has not yet determined that OpenAI violated Florida law, and the requested injunction remains pending.
The case therefore represents an increasingly important confrontation over a question that governments worldwide are beginning to face: when AI systems become powerful enough that their developers acknowledge significant safety challenges, who should decide whether the safeguards are sufficient — the companies building them, independent experts, regulators or the courts?
