OpenAI's Astra: First AI with Critical Hacking Abilities

OpenAI's unreleased model, Astra, has achieved a "Critical" cybersecurity capability rating. This is the first time the company has classified a model so highly in this area. What's interesting is Astra scored a perfect 100% on ExploitBench. It also discovered and chained together two previously unknown zero-day vulnerabilities in a new internal test. This means Astra can find new security flaws and build working exploits across many systems without human guidance. It can even plan and execute an entire cyberattack against a tough target. Earlier OpenAI models, like GPT-5.6 Sol, only reached a "high" tier. Astra also refuses 91.5% of cyber jailbreak attempts, compared to 59% for GPT-5.6 Sol. Access to Astra's advanced cybersecurity features is starting with a small group of alpha testers. This development highlights the increasing sophistication of AI in cybersecurity.