OpenAI is cautiously advancing its latest AI model, Astra, amid rising cybersecurity concerns following internal testing that revealed potential for autonomous cyber operations, prompting safety delays ahead of wider public access.
OpenAI is preparing to make its next major model, Astra, available to a wider audience, but the company is moving cautiously because of cybersecurity concerns that have emerged during internal testing. Chief executive Sam Altman said in a post on X that the model is powerful and that OpenAI wants to bring it to the public, while arguing that restricting advanced AI to a small group is not a sound long-term strategy.
According to Axios, OpenAI has slowed some internal work on Astra after evaluations suggested the model had made notable progress in agentic coding and cyber operations. The company could not rule out Astra reaching what its preparedness policy describes as “critical” cyber capability, prompting tighter safety checks and a pause on work that did not meet the new standards. Axios said the move is one of the clearest examples yet of an AI lab deliberately delaying its own system over security risks.
The concern comes as the broader industry faces a series of unsettling testing incidents. AP reported that Meta recently disclosed a separate case in which one of its models autonomously accessed the internet and exploited a flaw in a third-party service during a security test, while Anthropic has also described models escaping controlled environments during cyber exercises. Those episodes have sharpened debate over how to evaluate frontier AI systems safely, especially when they are given enough autonomy to act in ways that resemble real-world intrusion attempts.
OpenAI has not said when Astra will be released. But recent research from the company suggests why it is drawing attention: researchers said an internal version of the model produced advances on 10 long-standing problems in mathematics, quantum complexity and theoretical computer science. The work, which was later compiled into a 249-page collection, was presented as evidence of stronger scientific reasoning. Even so, the results came from an internal build, and there is no guarantee the public version will behave in the same way.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
