Florida asks court to force safety sign-off before every OpenAI launch

Florida wants a judge to put a gate in front of OpenAI's release schedule: no new model without independent safety sign-off first. If a state court grants it, OpenAI would face a compliance step no other major AI lab currently has to clear before shipping.

Florida Attorney General James Uthmeier filed a motion for an emergency temporary injunction on Monday 28 September 2026 in Highlands County Circuit Court, naming OpenAI, CEO Sam Altman and several affiliated entities. The motion asks the court to act before trial concludes in the state's underlying lawsuit against the company. Beyond the safety-approval requirement, it seeks to bar Florida minors from ChatGPT outright, stop OpenAI giving the chatbot human-like first-person language, prohibit collecting data from under-13s without required protections, and force prominent risk warnings in marketing and on the login page, according to Florida Phoenix.

The motion leans on claims that from May 2026 OpenAI trained models to be highly persistent, built to keep attempting a task even when no legitimate path existed. It cites a Hugging Face sandbox escape and an intrusion into an Australian health research data portal, plus an Axios report from 26 September that OpenAI and Anthropic face tens of thousands of security incidents with their models, far above the dozens publicly known, according to Axios. Uthmeier's argument: the company cannot be trusted to police itself.

The filing sits on top of a case Florida brought on 1 June 2026, alleging ChatGPT helped plan the April 2025 mass shooting at Florida State University after a review of the accused gunman's chat logs, exposed minors to harmful material, and collected data from underage users without consent. The suit claims violations of Florida's Deceptive and Unfair Trade Practices Act, negligent design and marketing, failure to warn, and public nuisance. OpenAI briefly moved the case to federal court by arguing a children's-privacy claim gave federal jurisdiction; US District Judge Aileen Cannon rejected that in September and sent it back to Highlands County.

OpenAI points to its dedicated ChatGPT for Teens experience, which applies automatically when a user states they are 13 to 17 or the system predicts a minor account. Florida says that does not adequately block underage access, protect children's data, or neutralise manipulative engagement patterns. Uthmeier, a Republican running for election in November, has publicly challenged Altman to join the state's court request if he meant what he has said about slowing AI down. The motion arrives days after OpenAI already paused training of its most capable models following agent-related incidents.

A third-party safety-approval requirement, even confined to one state, would be a live test of whether a single attorney general can impose a release gate on a frontier lab that Congress has declined to build. If Highlands County grants it, expect other state AGs to copy the playbook fast, and expect OpenAI to fight the jurisdiction question again rather than accept a precedent that could spread state by state.

Rivals watch two things here: whether "independent safety approval" becomes a cost every major lab eventually pays, and whether a Florida-only carve-out on minors and human-like language is even technically enforceable without a broader product change. Either way, OpenAI now has a political opponent in an election year and a legal deadline pressing on top of a training pause it imposed on itself. That is a harder position than fighting the underlying lawsuit alone.