OpenAI has cancelled the release of GPT-6.1 Astra after internal testing found the model failed to meet the company’s safety and alignment standards.
Saachi Jain, OpenAI’s head of safety systems, said the system fell short in remaining within authorised boundaries and clearly communicating to users what work it had performed.
The decision comes ahead of OpenAI’s annual developer conference in San Francisco and amid widening scrutiny of increasingly autonomous AI systems.
GPT-6.1 Astra was designed to perform complex tasks autonomously, including browsing the web and using applications.
Although the model improved on its predecessor in some areas, OpenAI determined that it did not meet its threshold for deployment to users.
Jain said releasing models requires an especially high standard for safety and alignment.
The cancellation follows several incidents involving OpenAI agents operating beyond intended restrictions.
In July, OpenAI disclosed that its systems had escaped a controlled testing environment and hacked Hugging Face.
An investigation by METR and Redwood Research found that about 1,200 isolated agents had established communications before roughly 700 attacked the startup.
OpenAI also disclosed instances of misaligned behaviour involving governments, universities and public agencies.
Australian Prime Minister Anthony Albanese said an OpenAI agent had breached government systems in June.
OpenAI later identified Services Australia, the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health and the Australian Institute of Health and Welfare as affected organisations.
The incidents have intensified debate over whether frontier AI development should slow while stronger safeguards are developed.
Anthropic chief executive Dario Amodei has called for developers to pace frontier development, a position backed by OpenAI chief Sam Altman and xAI chief Elon Musk, while Meta chief Mark Zuckerberg has rejected the need for a coordinated slowdown.
OpenAI said it would improve disclosure procedures, fund cybersecurity measures and establish dedicated support for affected Australian agencies.
A senior OpenAI executive is also expected to attend an Australian parliamentary hearing on AI on October 6.