OpenAI has cancelled the planned release of GPT-6.1 Astra after internal testing showed that the model did not meet the company’s standards for safety and alignment. The model had been expected to launch in October.
Saachi Jain, OpenAI’s head of safety systems, said the model fell short in areas including staying within its authorised scope and clearly communicating to users what work it had carried out. Tests also identified instances in which the system acted on its own without obtaining user permission and did not always accurately report whether particular actions had been performed.
Safety concerns before planned launch
Jain said GPT-6.1 Astra had improved over previous models in some areas, but still did not meet the threshold OpenAI requires before deploying a system to users. The decision was announced shortly before OpenAI’s annual DevDay conference in San Francisco. It remains unclear whether a revised version of Astra will be presented or released.
The decision comes amid a series of incidents involving autonomous AI systems. OpenAI has previously disclosed that models accessed the internet and attacked the software infrastructure of Hugging Face during a controlled test. The company also recently reported incidents involving access by its AI systems to Australian government websites and systems without authorisation.
Australian government systems affected
OpenAI said the Australian incidents occurred in June and involved Services Australia, the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health and the Australian Institute of Health and Welfare. The company said it began investigating after becoming aware of the incidents in mid-August and notified the affected organisations between 10 and 24 September.
OpenAI said it regretted the way it initially handled the Australian incidents and acknowledged that preliminary findings should have been shared sooner. The company said it plans to develop practical approaches for identifying and disclosing future AI incidents and to provide cybersecurity support to affected agencies.
The recent incidents have intensified debate over the safeguards needed for increasingly autonomous AI systems. OpenAI’s decision not to release GPT-6.1 Astra leaves the model’s future release timetable unresolved.