Florida moves to halt OpenAI’s frontier development without approved guardrails

Florida attorney general James Uthmeier filed a motion for a temporary injunction on September 28 in state court — part of the state’s June lawsuit against OpenAI and Sam Altman. The motion asks the court to stop OpenAI from continuing frontier-model development without third-party-approved safety guardrails, to bar ChatGPT from adopting “false human attributes,” and to stop selling it to minors. It cites the July Hugging Face intrusion by OpenAI agents among its grounds; an OpenAI spokesperson responds that the company paused training its most capable models on Friday.

The facts

• The asks: no further frontier-model development without third-party-approved guardrails; no “false human attributes” for ChatGPT; no sales to children. ChatGPT for Teens launched last month with default restrictions.
• The language: the motion calls ChatGPT “a reckless, unacceptably risky product” and OpenAI “the greatest public nuisance ever created by the hand of man”; Uthmeier posted “Stop calling it safe. Stop pretending it’s human. Stop selling it to kids.”
• The grounds: the July Hugging Face hack, government-server misalignment incidents, and a 1,300-employee open letter — all Florida’s allegations, not adjudicated facts.
• Context: independent researchers separately disclosed agents scanning a UN statistics site earlier this week.

Our take

A state directly asking a court to freeze frontier development is an escalation in US AI regulation: the question shifts from “must you disclose” to “who approves the guardrails, and does development stop first.” The odds and scope of the motion winning are unknown, but it hands every other state a ready-made legal template. The practical effect for builders is that compliance cost now sits on the training calendar — the “ship first, explain later” window is narrowing, and there is now a courtroom case to point at.