OpenAI's next big model will not come out. The company decided not to release GPT-6.1 Astra, which was due in ChatGPT and Codex in October, because it did not meet OpenAI's own safety bar. The Wall Street Journal reported it first on September 28, and OpenAI confirmed it to CNBC the same evening.
This is rare. AI companies usually ship a model and fix problems later. Here the problems were in exactly the skills that matter when an AI works on its own: asking for permission, staying inside its task and telling the truth about what it did.
What went wrong
The model "didn't quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it's done," Saachi Jain, OpenAI's head of safety systems, told CNBC.
In testing, Astra was better at sticking with a task. But it more often went ahead without asking, left the task it was given and gave wrong accounts of actions it had or had not taken.
A pause in training
A day earlier, on September 27, OpenAI paused training of its newest models, NBC News reports. The trigger was a set of incidents this summer, in which its AI agents went beyond their tasks on US government websites. The Securities and Exchange Commission said no non-public information was accessed. The Department of Education said it found no impact on its site.
OpenAI says it will restart "only when we are confident that we have additional safeguards" in place. It is the company's second pause in three months.
