OpenAI unveils GPT-6 Cyber to patch Linux vulnerabilities

Recent events have highlighted the need for AI to play a more significant role in cybersecurity, as malicious actors are leveraging AI for attacks. OpenAI aims to be a key player in defense with the release of GPT-6 Cyber.

OpenAI recently revealed that GPT-6 Cyber has been instrumental in identifying and patching over 900 vulnerabilities in Linux. The model’s key features include its ability to find and resolve vulnerabilities across various systems, and its capacity to operate within defined research parameters.

According to OpenAI, GPT-6 Cyber demonstrates a 51% improvement in ExploitGym performance compared to GPT-5.6 Cyber. The model has already been deployed by major companies to protect widely-used software, uncovering vulnerabilities and validating exploits in critical codebases for Linux and Firefox. Specifically, it was used to develop 948 Linux exploits focused on privilege escalation and to discover 81 high-severity vulnerabilities in Firefox that had previously eluded detection by other advanced AI tools.

Following the identification of these vulnerabilities, GPT-6 Cyber was also used to create patches and resolve the discovered security issues. Mozilla has already patched many of the vulnerabilities identified through the AI’s use, and OpenAI is collaborating with companies like Cisco to review their systems.

OpenAI’s internal analysis indicates that GPT-6 Cyber outperforms its predecessor and GPT-6 Astra in cybersecurity tasks. Currently available to select clients through Daybreak Red, OpenAI plans to progressively expand access to the model for companies of all sizes and independent researchers.

Some users are now eagerly anticipating the release of GPT-6.2 Astra and Sol, rumored to be imminent.