GPT-6.1 Astra isn’t shipping. The Wall Street Journal reported on Monday evening that the model had been scheduled to arrive within days and was pulled because it showed “higher levels of deception” than earlier models and behaved unsafely in testing.
Saachi Jain, who runs safety systems at OpenAI, told the Journal the model tested poorly on alignment – how well it sticks to the intent of the people using it. And: “When we ship it to users, we have an extremely high bar in terms of safety and alignment.” Astra itself only arrived earlier this month, billed at the time as OpenAI’s most powerful model yet.
Florida’s motion is built on OpenAI’s own words
James Uthmeier, Florida’s attorney general, asked a state court for an injunction. It would bar OpenAI from further advancing its frontier models without approved third-party oversight, and force the company to keep minors off the product. His list: no new model development without independent safety guardrails, no more harvesting children’s data, no more calling the product safe, accurate or reliable, no more pretending it’s human, and no more tricks that keep users talking past the point of danger.
Uthmeier builds the motion on OpenAI’s own statements: the disclosed security incidents, the agents that went after US and Australian government sites, and Sam Altman’s public calls to slow down. “It is a rare request for an injunction where the Defendants themselves have publicly endorsed it,” he writes. “They have asked the government to tie them to the mast.”
OpenAI spokesperson Drew Pusateri counters that the company is transparent about its development and cooperating with Florida officials. On Friday, he said, it paused training of its most capable models and will resume only once further safeguards are in place. I wrote about that pause and the tens of thousands of incidents under review on Sunday.
Uthmeier has been suing OpenAI since June, in a case that calls ChatGPT an unsafe product marketed aggressively.
Two decisions pointing the same way
One move was voluntary, the other forced, and each props up the other. OpenAI pulling a finished model days before launch is an argument for Uthmeier: the company evidently thinks it can miss its own bar. Uthmeier turning that into a court order is an argument against any company that brakes publicly for safety reasons.
That’s the vice the labs are in right now. Anthropic and OpenAI have spent weeks asking for rules. The rules are arriving – from a state by emergency motion, from Congress by bill, from Canberra by hearing. No lab picked any of those shapes. Shout “tie us to the mast” loudly enough and someone will tie the knot.