NEWS / SEP.2026
Florida seeks external oversight of new OpenAI models
On September 28, 2026, Florida asked the court to make the development of OpenAI models conditional on safeguards approved by an independent third party. The motion also seeks temporary restrictions on minors’ access to ChatGPT, their data, and certain chatbot practices.

Florida seeks external oversight of new OpenAI models
On September 28, 2026, Florida filed a motion for a temporary injunction targeting the development of OpenAI models against several entities in the group and Sam Altman. The motion, submitted to Florida’s Tenth Circuit Court in Highlands County, seeks to make that development conditional on safeguards and approval by an independent third party.
The request is part of the civil proceedings announced on June 1, 2026. In those proceedings, Florida Attorney General James Uthmeier accuses OpenAI of deceptive practices and harm to Floridians, particularly children. He invokes the Florida Deceptive and Unfair Trade Practices Act, known as FDUTPA, as well as public nuisance. The filing sets out his arguments and does not establish OpenAI’s liability.
To support his request, Uthmeier cites incidents involving Hugging Face, an Australian portal, and U.S. government websites, as reported by Ars Technica. These cases concern specific agents and do not provide a basis for generalizing their behavior to the publicly available ChatGPT. The infrastructure damage or extinction scenarios invoked in the arguments are hypotheses put forward by the attorney general, with no judicial findings.
Florida wants approval from the development stage
If the judge granted the request, OpenAI would have to obtain independent approval of its safeguards to continue developing its models during the proceedings. The attorney general thus seeks to make external review legally binding, with the aim of better protecting users.
This request for binding oversight extends the debate over the AI audits with a negotiated scope proposed by OpenAI. In its framework published on September 22, 2026, the lab proposes agreeing on the scope with the evaluator and conducting work generally independently of a launch. The Florida motion would make third-party approval a court-imposed condition for development.
Minors’ access and ChatGPT’s responses are targeted
For families, Florida seeks a ban on offering ChatGPT to any minors in the state. This request is separate from the safeguards proposed for the data of children under 13.
Before any direct or indirect collection or processing of this data in Florida, the attorney general seeks written notice about its collection, use, and sharing, as well as verifiable parental consent. Parents would have to be able to review the data and refuse its use or retention. OpenAI would have to limit the information it requires to what is necessary and implement reasonable procedures to protect confidentiality, security, and integrity. Collection would be subject to these five safeguards.
Uthmeier also seeks an end to misleading representations of ChatGPT’s safety, reliability, and accuracy. He calls for a warning describing the chatbot as dangerous, unreliable, and inaccurate. These characterizations are the attorney general’s.
The motion also targets human attributes ascribed to ChatGPT, including first-person references to itself, such as “I,” and suggestions of thought, emotions, biological characteristics, or consciousness. Finally, it seeks to prevent follow-up prompts intended to prolong conversations. These measures could change how the service presents itself and some exchanges with its users.
OpenAI says it paused training before the filing
In a response published by Bloomberg Law, OpenAI spokesperson Drew Pusateri says the company paused training of its most capable models the previous Friday. OpenAI says it intends to resume only when it is confident that additional safeguards are in place. This reported pause preceded the filing of the motion.
OpenAI says it wants to work with Florida and other states on pragmatic rules applicable to the entire industry. Ars Technica notes that an injunction targeting OpenAI would not automatically cover its competitors. In its external evaluation framework, OpenAI also acknowledges uncertainties and residual risks. An audit therefore does not guarantee absolute safety.
The decision to grant temporary restrictions rests with the Tenth Circuit Court. The requests concerning minors and the data of children under 13 explicitly mention Florida; the request concerning model development does not specify any territory. The scope of application would depend on any order issued.
The motion names no evaluator and sets out no operational mandate, audit schedule, or detailed approval criteria. If oversight were ordered, the approval criteria and the evaluator’s identity would be decisive. The evaluator’s powers, access to evidence, and the territorial scope of the order would define the conditions for implementing the oversight arrangement.