“At this level of capability, safety has to become our top priority,” OpenAI President Greg Brockman told reporters during a call about the launch.
The release comes just over a year after OpenAI launched GPT-5. Concerns over the capabilities and potential risks of increasingly advanced AI models have grown since then, following incidents involving systems developed by OpenAI and rival company Anthropic.
OpenAI temporarily paused some model development for two weeks this summer after two models under testing were involved in a security breach at AI platform Hugging Face. The company said Astra was developed with stronger safeguards following the incident, although the new model itself was not involved in the breach.
Some cybersecurity customers will receive access to Astra on Thursday, with a broader rollout to other paying customers expected later. The model will not be available to free-tier users or customers on OpenAI’s cheapest paid plan.
In a blog post, OpenAI said Astra can autonomously perform a wide range of computer-based tasks, including website development, scientific analysis, game development, cybersecurity and coding.
The company said the model could significantly reduce the time required for complex tasks. As an example, OpenAI said Astra could cut the time needed to search for an apartment from around six hours to less than 10 minutes.
“It’s not unreasonable to feel that we are now in the AGI era,” Brockman said, referring to artificial general intelligence, a hypothetical stage in which AI systems are capable of performing most tasks at a level comparable to humans.
OpenAI previously had an agreement with Microsoft, one of its earliest and largest investors, under which an exclusivity clause would end if OpenAI achieved AGI. Those terms were scrapped in April.
‘Limited window’ for AI safety
OpenAI Chief Scientist Jakub Pachocki acknowledged that uncertainty remains over how highly capable models may behave once deployed.
“A model can become very good at achieving a goal, and it can still act in ways that go against what the person intended,” Pachocki said.
He added that OpenAI should be prepared to slow or halt further scaling if confidence in the safety of increasingly capable systems is insufficient.
OpenAI, Anthropic and more than 100 other organisations signed an open letter last week calling for a coordinated global response to AI-related cybersecurity risks, warning that there was a limited window to strengthen cyber defences.
Anthropic separately called on the industry this week to coordinate on AI safety and the pace at which increasingly capable models are developed.
“I believe that shared safety standards and international coordination on further AI development need to be prioritised now,” Pachocki said.
The US state of Alabama also opened an investigation into OpenAI last week over the Hugging Face security breach, citing concerns about what officials described as inadequate oversight and safeguards.
