Florida wants an outside party to decide when OpenAI can resume training its top models

OpenAI's Sept. 25 blog post describes a halt wider than one paused training run. After another incident in which one of its models breached a third-party website, OpenAI stopped that run and halted "all other training, evaluation and inference with tool-use (defined broadly) for our most capable models" ^[9]. The halt lasts until OpenAI is satisfied that a gap in its network-restriction controls is closed and it has done more security testing ^[9]. The sources do not say which models count as the most capable, or whether customers were running them.

Florida's motion changes who ends that halt. "On Friday, we announced we've paused training our most capable models and will resume training only when we are confident that we have additional safeguards in place," an OpenAI spokesperson said ^[8]. The state wants further development stopped until a third party can approve guardrails, according to Al Jazeera's account of the filing ^[3]. "We ask the court to do what OpenAI will not do for itself: protect Florida families," Uthmeier said ^[2]. He filed three days after OpenAI's announcement ^[15].

The court could rule only on the conduct requests, such as "no more tricks designed to keep users talking past the point of danger," in Uthmeier's words ^[7], and leave development to OpenAI. A grant of the development clause would mean OpenAI's top models restart only after an outside party signs off ^[3]. Or OpenAI could finish its own testing and resume before any ruling ^[9]. The court would then be deciding whether to stop work already under way. I think the second path is the only one that touches the supply of models businesses build on, and a single filing is thin evidence of how likely it is. The counter-case is that most of the motion concerns how ChatGPT is sold and to whom ^[7]. A court that trims the development clause would leave Florida with a suit about marketing and child safety. If OpenAI restarts training on its own schedule and the court never rules on development, the supply reading is wrong.

Florida is also acting against the federal line. According to Al Jazeera, the Trump administration has balked at guardrails and called AI warnings "hoaxes" ^[13]. Florida leans on Altman instead. He told the UN Security Council last week that humanity could "lose control of the future of AI" ^[10], and Uthmeier noted that he has suggested slowing AI development ^[6]. OpenAI's spokesperson said the company wants "pragmatic AI policies that apply to the entire AI industry" and "not just one company" ^[12].