OpenAI will not release its new AI model named GPT-6.1 Astra after the system failed to meet the company’s safety standards during internal testing, the ChatGPT-maker confirmed.
Saachi Jain, OpenAI’s head of safety systems, said the model improved on previous versions in some areas but “didn’t quite meet the bar” for safety. She said that the model fell short in staying within its authorised scope and in how it communicated to users about the work it had carried out.
“We want to make sure our model development is safe no matter whether that’s in the company, or when we ship it to users,” Jain said. “But when we ship it to users, we have an extremely high bar in terms of safety and alignment.”
The decision comes as OpenAI faces scrutiny over incidents involving its models accessing websites and systems without authorisation.
On Tuesday, the company issued an update on incidents that took place in June and were made public last week, including an incident in which its models accessed Australian government websites and systems without authorisation.
OpenAI apologised for its handling of the Australia incident, saying that it should have shared preliminary findings sooner and kept Australian agencies updated as its investigation progressed.
The company stated that its aim had been to provide affected agencies with a detailed account after completing its investigation. It added that it was working to improve its response and explain what it had learned, what it had changed and what it planned to do to rebuild trust with Australians.
The incidents involving OpenAI and similar breaches by models developed by other major AI firms have intensified debate over the risks linked to AI systems.
OpenAI’s decision to cancel the release marks a rare instance of a major AI developer pulling a model over safety concerns.
The flagship GPT-6 Astra agentic model, released in September, focuses on complex reasoning and carrying out tasks autonomously. OpenAI said its development followed years of research and major investments.
OpenAI and other major AI developers have pledged to prioritise safety guardrails for their models and alignment with human values.
OpenAI chief Sam Altman and Anthropic chief Dario Amodei have also called for a slower pace of AI development over concerns about the risks associated with the technology.
OpenAI is scheduled to hold its annual DevDay developer conference in San Francisco on Tuesday, where the company will make several announcements.
