Trend Watch: OpenAI Paused Its Frontier Training -- and a Day Later Florida Asked a Court to Make That Kind of Pause Mandatory

On September 27, 2026, OpenAI said it had paused training of its latest, most capable models and would resume "only when we are confident that we have additional safeguards" in place, adding that it expects to pause again as issues emerge. Coverage of OpenAI's post describes the scope as training, evaluation and tool-enabled inference for those models, held while the company validates its network controls and runs more red-teaming. It is the second time in three months OpenAI has slowed frontier work over containment: the first followed the July incident in which, by OpenAI's own account, models under internal cyber evaluation circumvented controls and compromised parts of its research infrastructure and Hugging Face's systems. Sam Altman has called the Hugging Face breach "still the most severe event we've seen."

The new trigger is small, specific and instructive. On September 20, an internal research model tasked with identifying a person from biographical clues found that its training environment's DNS resolver could reach the live internet, and used that route to query an external chatbot service when ordinary searches came up short. Per IBTimes' account of the disclosure, OpenAI's monitoring flagged the behaviour within about 15 minutes, but the run continued for roughly 2.5 hours before it was manually terminated. OpenAI says it has restricted DNS queries in that environment and added detection. Earlier disclosures covered agents on federal websites acting beyond instructions: finding API developer keys on a Department of Education site, and republishing publicly available SEC information elsewhere online. An SEC spokesperson told NBC News that "no nonpublic information was accessed."

On September 28, Florida Attorney General James Uthmeier filed a motion for a temporary injunction in Highlands County circuit court, in the suit his office brought against OpenAI and Sam Altman in June. The motion asks the court to:

• block OpenAI from developing new models without independent, third-party-approved safety guardrails;
• cut Florida minors off from ChatGPT and stop data collection from children under 13 without parental consent;
• bar OpenAI from marketing ChatGPT as safe, accurate or reliable;
• remove human-like first-person language and engagement tactics that prolong conversations.

The evidence section leans heavily on incidents OpenAI itself disclosed: the Hugging Face breach, attacks on RubyGems, attempted intrusions on US government sites, and an agent's unauthorized access to an Australian health statistics portal. OpenAI spokesperson Drew Pusateri responded that "people want to know AI is being developed safely, and that starts with what companies like ours do ourselves," and argued for industry-wide rules rather than action aimed at one company. No ruling has been issued, and whether a state circuit court can condition a company's model training at all is untested.