OpenAI's GPT-6 Astra Achieves Critical Offensive Cyber Capabilities

OpenAI's new frontier model, GPT-6 Astra, has surpassed the company's internal "critical" threshold for offensive cyber capabilities. The model achieved 100% on ExploitBench and 42.4% on ExploitGym, benchmarks for developing and chaining novel exploits. This development raises significant concerns regarding AI safety and the potential for autonomous exploit generation, highlighting the urgent need for robust AI governance.