SafetyOpenAI
OpenAI released GPT-6 Cyber to approved defenders and rated it Critical for cybersecurity
948Linux kernel privilege escalation exploits OpenAI says it built, against 166 for GPT-5.6 Sol
OpenAI released GPT-6 Cyber on 5 November to approved Daybreak Red customers through the API and Codex. It scores 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber, and OpenAI rates it Critical for cybersecurity under its Preparedness Framework, up from High. OpenAI says it found 81 new high severity Firefox flaws. In a test of staying in scope it avoided out-of-scope services 99.93% of the time without deployment prompts. Every figure is OpenAI's own.