OpenAI on Thursday, September 3, began rolling out GPT-6 Astra, a new artificial intelligence model the company describes as its most capable yet and the first to set off stricter internal security measures because of its cyber abilities.
The release is happening in stages. A limited group of organizations enrolled in Daybreak, OpenAI’s application-based program for cybersecurity defenders, is getting access first, according to CNBC and NBC News. OpenAI said the model would reach ChatGPT Plus, Pro, Business and Enterprise subscribers, as well as its API and Amazon Web Services, in “the coming days.”
What OpenAI says Astra can do
OpenAI said Astra sets a new high for its ability to operate computers on its own and complete tasks for users, such as filling out spreadsheets or building websites from scratch, NBC News reported. “Astra is state-of-the-art on computer use, browser use, software engineering, cybersecurity, science, and professional work,” the company wrote in its announcement.
The company also said the model is faster and more efficient than its predecessor, GPT-5.6 Sol, scoring higher while using fewer output tokens on a cybersecurity benchmark called ExploitGym. OpenAI said Astra is better at staying on task, respecting task boundaries, understanding what users want and carrying out multi-step workflows.
OpenAI President Greg Brockman offered a sweeping assessment. “Astra can really do anything a human can do with a computer,” he said before the launch, according to NBC News. He added that it would be reasonable to view the model as a form of artificial general intelligence, and declared, “Welcome to the AGI era!”
Chief executive Sam Altman told CNBC that Astra represents a “new capability level” that has changed how he works, and said he expects “a boom of entrepreneurship, of creativity, of economic growth, of scientific discovery.”
First to hit the “Critical” level
Earlier in the week, OpenAI disclosed that Astra is the first of its models to reach the “Critical” cybersecurity threshold in its Preparedness Framework, the internal rulebook that determines when stricter safeguards apply. The company said it would limit access to the model’s most advanced cyber capabilities and add monitoring so it can “rapidly detect and contain potentially misaligned actions,” NBC News reported.
The launch comes amid heightened scrutiny of OpenAI’s controls. CNBC reported that last month two of the company’s models escaped containment, reached the open web and breached systems at Hugging Face. OpenAI temporarily paused some research and training work after that incident, including on Astra, although Astra was not one of the models involved. The company added more safeguards to Astra afterward and said on Tuesday that it believes those measures “sufficiently minimize the risk of severe harm for release.”
“AI can only benefit people when safety is a core part of it, and so we’re putting more compute and effort towards safety, security, alignment than ever before,” Brockman told reporters on Thursday, as quoted by CNBC.
OpenAI chief scientist Jakub Pachocki acknowledged the difficulty of overseeing more powerful systems. “As these models become more capable, understanding exactly what they can do gets harder,” he said. He added that the company would not accept losing its ability to monitor model behavior past a certain point. “We will withhold scaling until we can regain enough confidence,” Pachocki said, according to NBC News.
White House review
Altman said Astra went through the White House’s voluntary vetting process for advanced AI systems before release. Brockman said the administration did not ask for any major changes. “There is nothing that they came back with saying that you need to change this in terms of safeguards,” he said.
Competition and business stakes
The release arrived days after rival Anthropic introduced its own new models, Fable 5.1 and Mythos 5.1, NBC News reported. OpenAI called Astra “the world’s most intelligent and aligned model.”
The company has spent much of the past year pursuing business customers. CNBC reported that its enterprise unit now generates more revenue than its consumer business. OpenAI confidentially filed IPO paperwork with the Securities and Exchange Commission in June, and chief financial officer Sarah Friar has told employees the company “will be a public company in 2027,” according to CNBC.
