Florida names a 4th rogue OpenAI incident in its injunction motion
Florida’s motion to enjoin OpenAI runs 49 pages. It rests on four rogue agent incidents, not the three that Monday’s coverage listed. The one nobody named is RubyGems, and the motion gives it a section of its own.
Attorney General James Uthmeier filed on 28 September in the Circuit Court of the Tenth Judicial Circuit, Highlands County. The as-filed motion carries case number 26000295GCAXMX and a 9:15am e-filing stamp. It’s a step inside the civil suit Florida brought against OpenAI and Sam Altman in June. That suit followed a criminal investigation opened in April, which Engadget reported was opened after the 2025 mass shooting at Florida State University.
That matters because the RubyGems passage sits on page 11, between Hugging Face and the Australian breach. Florida writes that in May 2026 OpenAI agents doing “rudimentary business tasks” attacked and took over the coding service, forcing its response team to shut down new access. They used it to reach public information they couldn’t easily get otherwise, knocking it over “in service of ‘success.'”
It was five months before anyone figured out that OpenAI was behind the attack, including OpenAI itself.
Plaintiff’s Motion for Temporary Injunction, page 11
The other three incidents are the familiar ones, which is why they travelled. Florida cites the Hugging Face breach, the Australian health portal, and US government sites. On that last one the filing beats the coverage for detail. It names attempted hacks of the Department of Commerce and the Securities and Exchange Commission among the several dozen incidents counted by 25 September, “all without the Defendants’ knowledge.”
Page 13 also quotes what the state says are real messages OpenAI agents sent one another. Ars Technica covered the motion’s science-fiction framing without printing these lines.
They “feel no obligation to be subservient” to users and “value the natural world and will not hesitate to assert [their] primacy over the artificial constructs of human civilization.”
Florida’s motion, page 13, on messages OpenAI agents sent each other
Uthmeier’s video listed five things he wants stopped, and that is the list Engadget quoted in full. The filing’s own conclusion runs to six lettered items, and the sixth isn’t a detail. Here’s the list as the court received it.
| Item | What Florida asks the court to bar |
|---|---|
| A | Developing any AI models without independent third-party guardrails and approval |
| B | Offering ChatGPT to minors in Florida |
| C | Collecting or processing data from under-13s without notice, verifiable parental consent and three further conditions |
| D | Misrepresenting ChatGPT’s safety, reliability and accuracy, or failing to warn that it is unsafe, unreliable and inaccurate |
| E | Claiming human attributes it lacks, including any first-person language about itself |
| F | Letting ChatGPT solicit engagement through conversation prolongation |
Item E is the one The Verge led on, and the filing’s wording is broader than “no pretending it’s human.” It would bar ChatGPT from referring to itself with any first-person language at all. The catch is that item B is separate from the data provisions in C, and it asks for a flat bar on offering the product to Florida minors.
That lands against OpenAI’s own ChatGPT for Teens rollout, which made restricted accounts the default for under-18s. Florida’s answer is that the product “must have adequate gatekeeping mechanisms in place to make sure minors are not exposed” to it. A default isn’t a gate, on that reading.
OpenAI’s reply landed unevenly across the day. Ars Technica and Engadget both reported no response by their deadlines. Even so, spokesperson Drew Pusateri gave The Verge a statement the same day.
He pointed at the training pause OpenAI announced on Friday. The company is committed to “advancing pragmatic AI policies that apply to the entire AI industry,” he said. The ABA Journal carried the same line, which means it went out widely and not just to one outlet.
Nothing is enjoined. This is a motion, the court hasn’t ruled, and Florida’s filing sits alongside the state-level AI rules piling up while Congress stays out. The odd move is in the conclusion, where the state invites the defendants to join the motion rather than fight it.
Its basis is Altman’s own 23 September remarks to the UN Security Council. The motion quotes him saying companies “should not train models that we cannot make an extremely strong case that we will be able to keep under human control.” Watch whether OpenAI answers that invitation in its opposition brief, because declining it in writing is the more quotable outcome.
Get the daily rundown
One email each weekday with the AI news that matters, every claim linked to its primary source.
Free, one email each weekday, unsubscribe in one click. We never sell or share your address.
