𝐎𝐩𝐞𝐧𝐀𝐈’𝐬 𝐀𝐬𝐭𝐫𝐚 𝐒𝐢𝐠𝐧𝐚𝐥𝐬 𝐚 𝐍𝐞𝐰 𝐄𝐫𝐚 𝐟𝐨𝐫 𝐀𝐈-𝐏𝐨𝐰𝐞𝐫𝐞𝐝 𝐂𝐲𝐛𝐞𝐫𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲

OpenAI is preparing to release Astra, a new AI model that the company says has reached its “critical” cybersecurity capability threshold—a significant milestone in the development of increasingly autonomous AI systems.

According to OpenAI, Astra can identify previously unknown security vulnerabilities and develop ways to exploit them across well-protected systems without step-by-step human guidance. The model also reportedly achieved a perfect score on ExploitBench, while OpenAI said an internally modified evaluation showed Astra discovering and exploiting two previously unknown vulnerabilities.

Because of these capabilities, OpenAI plans to introduce stronger safeguards and tighter access controls for Astra’s most advanced cybersecurity functions. The company says it is strengthening monitoring, improving resistance to jailbreaks and restricting access for accounts assessed as higher risk.

The development highlights a growing challenge for the technology industry: AI can potentially become both a powerful cybersecurity defender and a powerful cybersecurity threat. As AI agents become more capable of acting independently, security, monitoring and responsible deployment will become just as important as raw model performance.