OpenAI has released GPT-6 Cyber, a security-focused model it rates "Critical" for cybersecurity capability under its own Preparedness Framework, the highest level, up from "High" for the previous version. In testing, the model built working root exploits for 948 separate Linux kernel flaws and found 81 previously unknown high-severity bugs in Firefox. It is not public: only approved customers in OpenAI’s Daybreak Red programme can use it.
What did GPT-6 Cyber find in Linux?
OpenAI ran an automated pipeline over the Linux kernel, the core of most servers and Android phones. Per BleepingComputer, the model flagged 2,308 potential flaws, reproduced 1,724 of them, and built local privilege escalation exploits for 948 distinct bugs. In plain terms, each of those exploits lets someone who already has a basic foothold on a Linux machine become its all-powerful "root" user. OpenAI says the pipeline also drafted fixes, and that humans reviewed findings before they were disclosed to kernel maintainers.
In Firefox, the model found 81 high-severity vulnerabilities and helped develop patches, many of which Mozilla has already shipped. On the ExploitGym benchmark it exploited 48.5% of test vulnerabilities, against 32.1% for GPT-5.6 Cyber.
Why is it rated Critical?
Because a model that can find and weaponise hundreds of deep operating system bugs on its own would be extremely dangerous in the wrong hands. That is the whole tension of AI security research in 2026: the capability that lets defenders fix 948 kernel bugs is the same one that would let an attacker exploit them. OpenAI’s own framing is blunt: "we have a limited window to use frontier models to secure software."
Who can use it?
Only vetted security teams. Access is limited to approved customers in Daybreak Red, OpenAI’s programme for authorised penetration testing and exploit development, with plans to widen it to all Daybreak Red customers. A separate tier, Daybreak Blue, gives verified defenders more room to investigate bugs and validate fixes. It is the same gated approach Anthropic took when it split its cyber access into three verified tiers, and the opposite of Mistral’s choice to release Large 4 with open weights.
What does this mean for everyone else?
Expect a wave of patches. If AI can surface hundreds of kernel and browser bugs in one run, the next few months will bring more security updates than usual for Linux distributions, Android and browsers. The practical advice is simple: turn on automatic updates and install them promptly. October showed what happens when attackers move first: PHP2Shell was exploited before its patch existed. See our Black October roundup for the wider picture.
