OpenAI releases GPT-6 Cyber to select Daybreak Red customers after reporting 948 Linux kernel privilege-escalation bugs

GPT-6 Cyber is a specialized version of GPT-6 Astra. It is available through the API and Codex to a small group of Daybreak Red customers, and OpenAI plans to widen Red access later. OpenAI rates the model as having Critical cybersecurity capability under its Preparedness Framework. OpenAI reports that the model scored 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber. That figure is vendor-reported, and the benchmark's task design is not described in public. OpenAI also says the model strayed out of scope far less often than its predecessor. [1] [3] [9] [10] [11]

OpenAI describes a five-stage kernel pipeline: discovery, sandboxed validation, exploitability assessment, patch generation and human review. In a mid-October run with an unreleased cyber model, it narrowed 2,308 candidates down to 948 unique privilege-escalation bugs. The run cost about $1.21 million, or roughly $1,279 per exploit. A July run with GPT-5.6 Sol found 166 such bugs for about $514,000. [7] [8]

Finding bugs is now moving faster than fixing them. By late October, Patch the Planet had logged 1,995 issues, written 679 patches and had 466 patches accepted upstream. Fewer than a quarter of issues have been remediated. Separately, a hobbyist using the Blue tier found a severe Ubuntu bug in about two hours. [1] [10] [13]

Access is still uneven. OpenAI has committed $1 billion in subsidized Daybreak credits for utilities, local governments, community banks and open-source projects. However, Red remains the most restricted tier, and some defenders say Blue was recently weakened. They argue that open-weight models with safeguards removed leave defenders at a disadvantage. [1] [5] [6]