OpenAI has introduced GPT-6 Cyber, an AI model designed for vulnerability discovery and cyberattack simulation. In a controlled environment, it successfully completed 95% of exploitation tasks in GitLab, PostgreSQL, and Jenkins, and achieved a 48.5% success rate in the ExploitGym test, compared to 32.1% for GPT-5.6 Cyber. Access to GPT-6 Cyber is currently limited to approved specialists.
1,724 Linux Vulnerabilities Reproduced, 948 Working Exploits Developed
The model reproduced 1,724 vulnerabilities in Linux and developed working exploits for 948 of them—five times more than the previous model in similar tests.
Firefox: 81 New High-Severity Vulnerabilities
In Firefox, 81 previously unknown high-severity vulnerabilities were discovered. An additional 30 known issues were found to be more serious than previously assessed.
Energy Testbed: Two New Software Vulnerabilities
During tests simulating a power plant, the AI identified two previously unknown vulnerabilities in software related to protective equipment settings and switchgear management.
Model Assigned Critical Capability Level
OpenAI has assigned GPT-6 Cyber a critical capability level for cyberattacks; access to the model is restricted to approved specialists.
