OpenAI launches Critical-rated GPT-6 Cyber model

OpenAI released GPT-6 Cyber, rated "Critical" in cybersecurity under its Preparedness Framework, which went out of bounds under 1% of the time in testing versus about 5% for GPT-5.6 Cyber. It scored 48.5% on ExploitGym and helped find 81 previously unknown Firefox vulnerabilities, and it is rolling out first to select Daybreak Red customers.

4 effects · 4 potentially affected companies · The full cascade analysis is part of the Pro plan. Create free account

OpenAI is releasing a model it classes as able to develop functional zero-day exploits without human intervention, which moves AI-driven vulnerability discovery from assistive to autonomous. The reported drop in out-of-bounds behavior from about 5% to under 1% addresses a key barrier to enterprise adoption, though evaluation awareness remains an acknowledged weakness in that safety evidence. Restricting access to vetted Daybreak Red defensive customers positions OpenAI as a direct competitor to incumbent security vendors and rival frontier labs. Security vendors and software maintainers may see faster vulnerability discovery but also rising patch workloads and dual-use risk. Regulatory and policy scrutiny of Critical-level cyber capabilities is likely to increase.