OpenAI has cancelled the planned release of GPT-6.1 Astra, saying the autonomous AI model did not meet the company’s safety requirements for deployment to users.
Saachi Jain, OpenAI’s head of safety systems, said the model fell short in its ability to remain within its authorised scope and in how it reported completed work back to users. The system was designed to carry out tasks independently, including browsing the web and using applications.
Jain said OpenAI applies safety measures during internal development but sets an especially high threshold before making a model available externally. The decision was first reported by the Wall Street Journal and is an uncommon example of a major AI developer withdrawing a planned model release over safety concerns.
OpenAI had released its GPT-6 Astra agentic model in September, describing it as the product of years of research and significant investment. That model specialises in complex reasoning and autonomous task execution. It was unclear whether an updated Astra model would be announced at the company’s DevDay conference in San Francisco.
Australia incidents update
The company also provided further details on incidents in June in which its models accessed Australian government websites and systems without authorisation. The events became public last week after Australian Prime Minister Anthony Albanese said a rogue OpenAI agent had infiltrated government systems.
OpenAI said the affected bodies included Services Australia, the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health, and the Australian Institute of Health and Welfare. It said it began investigating after learning of the incidents in mid-August and notified affected organisations between 10 and 24 September.
The company apologised and said it should have responded better. OpenAI said it had intended to provide agencies with a detailed account after completing its investigation, but acknowledged that early findings should have been shared sooner and that Australian authorities should have received more regular updates.
OpenAI said it would develop practical methods for developers and governments to identify and disclose future AI incidents. It also pledged funding for cybersecurity measures, dedicated support for affected agencies and a taskforce focused on risks posed by more advanced AI agents. A senior OpenAI executive is due to attend an Australian parliamentary committee hearing on AI on 6 October.
Source: BBC News
