Astra Becomes First AI Model to Reach Critical Hacking Threshold

OpenAI's Astra model has reached new heights in cybersecurity capability, earning its first 'Critical' rating under the company's Preparedness Framework. This milestone was announced on September 1, and marks a significant advancement for AI-powered security testing.

Astra demonstrated its prowess by scoring a perfect 100% on ExploitBench, a benchmark that tests models' ability to turn known software vulnerabilities into functioning exploits. In an internal test built from V8 browser vulnerabilities disclosed between June and August, Astra beat GPT-5.6 Sol in arbitrary code-execution rates, using far fewer output tokens.

But the most impressive feat was yet to come: Astra found and chained together two previously unknown zero-day vulnerabilities on its own, without human guidance. This shows a level of sophistication that raises concerns about potential security risks if left unchecked.