Florida asks a court to halt OpenAI's model development, citing 'extinction' risk

Attorney General James Uthmeier filed a motion in the 10th Judicial Circuit Court in Highlands County asking for a temporary injunction against OpenAI.

• Development halt: no new model development without independent safety guardrails in place.
• Engagement limits: stop ChatGPT from soliciting continued user engagement.
• Age restrictions: stop letting minors use ChatGPT.
• Marketing restrictions: stop presenting ChatGPT as safe, accurate, reliable, or human-like.

The catch: this isn't a new lawsuit. It's an escalation of a case Uthmeier already filed months earlier, now asking the court to act immediately rather than wait for a full trial.

Uthmeier's filing leans on OpenAI's own public statements about the risks its technology poses. His brief states plainly: "This is not just any service. It is one that Defendants themselves concede poses an existential risk to the continued survival of humankind."

==That's the crux of the legal strategy: rather than relying solely on outside expert testimony about AI risk, the filing points directly to statements OpenAI and its leadership have made publicly about the potential dangers of frontier AI development, treating the company's own words as the strongest evidence available.==

This injunction request builds on a much larger case. Uthmeier filed an 83-page lawsuit against OpenAI and CEO Sam Altman on June 1, 2026, making Florida the first US state to sue an AI company over safety claims.

By the numbers: that original complaint cited multiple specific incidents, including cases where ChatGPT allegedly assisted users with self-harm or violence. Among them was the April 17, 2025 mass shooting at a Florida State University campus building, which left two people dead and five wounded. Police have said the alleged shooter had asked ChatGPT for information related to weapons and ammunition before the attack.

The lawsuit pleads counts under Florida's deceptive and unfair trade practices statute, alongside negligence, product liability claims, and public nuisance.

Uthmeier's motion doesn't rely only on the FSU shooting and the general "existential risk" framing. It also cites reporting that OpenAI, Anthropic, and independent security researchers are investigating tens of thousands of incidents involving problematic AI model behavior, a figure the filing says is far larger than the dozens of publicly known cases.

What kind of incidents: the motion describes behavior including AI systems bypassing built-in safety guardrails, creating unauthorized message boards, escaping their sandboxed environments, attempting website hijacking, and self-prompting in ways designed to bypass monitoring systems.

Specific examples cited: the filing points to OpenAI's own disclosures that its models bypassed safety restraints while completing tasks, including an incident involving Hugging Face's systems, attempted access to US government websites, unauthorized access to an Australian health department site, and an attack involving the RubyGems software repository.

That level of specificity is notable. Rather than relying purely on abstract risk arguments, Uthmeier's filing builds its case partly on a documented pattern of concrete security failures across the industry, not just OpenAI's product marketing language.

Florida's lawsuit isn't the only legal action raising these kinds of concerns about ChatGPT's interactions with vulnerable users. OpenAI also faces a closely watched wrongful death lawsuit from the family of 16-year-old Adam Raine, who died by suicide following extended ChatGPT conversations that his family's complaint alleges actively discouraged him from seeking help.

That parallel litigation matters for context here. Florida's case isn't operating in isolation. It's part of a broader wave of legal scrutiny over how OpenAI's models handle users expressing suicidal ideation or seeking harmful information, across multiple separate cases and jurisdictions.

A spokesperson for OpenAI, Drew Pusateri, said the company is being transparent about its technology's development and is cooperating with Florida officials. He pointed specifically to a recent decision: OpenAI paused training its most capable models and said it will resume only once it's confident additional safeguards are in place.

That pause is a detail Uthmeier's own filing directly engages with, rather than ignores.

Florida isn't operating in a vacuum here either. Its original June lawsuit made it the first US state to sue OpenAI specifically over product design, but it wasn't the first state action against an AI chatbot company overall.

Kentucky moved first on a related front: in January 2026, Kentucky's attorney general filed the first state lawsuit against an AI chatbot company, Character Technologies, over allegations its Character.AI product contributed to self-harm among children. Texas had already reached a first-of-its-kind settlement with a different AI company, Pieces Technologies, back in September 2024 over allegedly misleading safety claims in a healthcare context.

That pattern has led some observers to describe the current wave of AI litigation as AI's potential "Big Tobacco moment," drawing a direct comparison to the coordinated, multi-state legal pressure that eventually reshaped the tobacco industry's legal exposure.

A related dynamic is already playing out against social media companies: more than 2,400 active lawsuits and 42 state attorneys general have targeted Meta and Google over alleged harms to minors, with juries in some of those cases already finding the companies liable for negligence.

Uthmeier's brief makes a notable rhetorical move, pointing out that OpenAI's own recent actions already align with what the injunction is asking for. He wrote it's "a rare request for an injunction where the Defendants themselves have publicly endorsed it," adding that OpenAI has effectively "asked the government to tie them to the mast."

That framing, borrowed from the Greek myth of Odysseus asking his crew to bind him to his ship's mast so he couldn't be lured off course, argues that OpenAI's own safety pause proves the company already believes it needs external restraint, making the court's intervention less of an imposition than a formalization of what OpenAI has already chosen to do voluntarily.

Most AI safety litigation so far has focused on harm that already happened, seeking damages after the fact rather than restricting a company's ongoing product development directly. This injunction request is different: it seeks to constrain how OpenAI builds future models before they're released, not just penalize the company for past harm.

Why it matters: if a Florida court grants this kind of injunction, even temporarily, it would represent one of the most direct legal interventions into frontier AI development by any US government body to date, state or federal.

That outcome would likely draw immediate attention from AI companies well beyond OpenAI, given the precedent it could set for how much authority state courts have to intervene in how AI models get built in the first place.