OpenAI's GPT-6 Astra Reaches "Cyber Critical" Threshold, Discovers Zero-Day Vulnerabilities

OpenAI's latest model, GPT-6 Astra, has achieved a "cyber critical" status, demonstrating advanced cybersecurity capabilities. During testing, Astra scored 100% on ExploitBench and a 42.4% success rate on ExploitGym, a benchmark for novel exploit chains. Notably, Astra independently discovered and chained two previously unknown zero-day flaws against high-severity vulnerabilities. This milestone, according to OpenAI CEO Sam Altman, required new safeguards for Astra's release, highlighting the evolving landscape of AI-driven cyberattacks.