OpenAI has launched GPT-6 Astra, its latest artificial intelligence model, with major advances in computer use, coding, scientific research and cybersecurity.
The company says Astra is its most capable broadly deployed model to date and the first OpenAI model to reach the “Critical” cybersecurity capability threshold under its Preparedness Framework.
The classification reflects Astra’s ability, under the right conditions, to identify previously unknown vulnerabilities and develop working exploit chains against hardened systems with limited or no human intervention.
OpenAI said internal testing showed Astra performing substantially better than its predecessor, GPT-5.6 Sol, on vulnerability identification and exploit development. On ExploitBench, a benchmark designed to test exploit-development capabilities, Astra achieved a 100% score.
The company also said that during testing on an internal benchmark containing recently disclosed high-severity vulnerabilities, Astra discovered and used two previously unknown vulnerabilities as part of an exploit chain. OpenAI said it is working to disclose those vulnerabilities to the relevant maintainers.
Advanced computer use
Beyond cybersecurity, Astra is designed to perform complex tasks directly through computer interfaces.
The model can navigate websites and software, complete online forms, work with spreadsheets and documents, conduct research, write and test software, and handle multistep professional workflows.
OpenAI said Astra can also produce documents, spreadsheets and presentations and assist with scientific and technical work. The company describes these capabilities as a significant expansion of AI systems from answering questions to completing tasks.
OpenAI President Greg Brockman has described the launch as a potential beginning of the AGI era, referring to artificial general intelligence. The company has positioned Astra as a major step toward AI systems capable of carrying out a broad range of economically valuable work.
Cybersecurity access will be restricted
Because of Astra’s cyber capabilities, OpenAI is limiting access to some of its most advanced cybersecurity functions.
The company said advanced cybersecurity work will initially be available to a limited group of testers, with access through its Daybreak program expected to expand defensive use of the technology.
OpenAI said additional safety checks may slow, pause or stop some legitimate cybersecurity tasks as the company attempts to prevent the model from being used to cause serious harm.
The company has also increased monitoring and security measures following a separate incident involving OpenAI systems and Hugging Face. OpenAI has said Astra itself was not responsible for that incident.
Safety concerns remain
The launch comes as the AI industry faces growing questions over the safety of increasingly autonomous models.
Astra’s ability to perform complex tasks with less human intervention creates potential benefits for developers, researchers and cybersecurity professionals, but it also increases the consequences if the system behaves unexpectedly or is misused.
OpenAI said Astra is more likely than GPT-5.6 Sol to follow explicit safety restrictions. At the same time, the company acknowledges that stronger safeguards are necessary because of the model’s increased capabilities.
OpenAI steps up AI competition
The Astra launch also increases competition among leading AI companies, particularly OpenAI and Anthropic, as they develop models capable of performing increasingly complex professional tasks.
OpenAI is targeting enterprise users with Astra’s improvements in software engineering, computer use, research and automation. The model is being rolled out gradually, with broader availability expected following the initial release.
For OpenAI, Astra represents both a technological milestone and a new safety challenge: the model can perform substantially more sophisticated work, but its cybersecurity capabilities have reached a level that the company itself classifies as Critical.
The rollout will therefore test whether OpenAI can make increasingly autonomous AI systems useful at scale while maintaining sufficient safeguards against misuse.
