OpenAI says GPT-6 Cyber has found thousands of exploits

In a blog post published Thursday, OpenAI described the model as delivering “major gains” on key cybersecurity benchmarks compared with GPT-5.6 Cyber. The announcement follows reports that the model found and patched 948 local privilege-escalation vulnerabilities in Linux and identified 81 previously unknown “high-severity” vulnerabilities in Firefox, despite extensive prior testing with frontier models.

OpenAI is restricting access through Daybreak, its cybersecurity program for verified individuals and organizations. GPT-6 Cyber is initially available only to a “select group” of approved Daybreak Red customers. Daybreak Blue is for defensive work; Red is for advanced penetration testing and security research. OpenAI said access will expand to all Daybreak Red customers “soon.”

The company said it aims to “dramatically” broaden Daybreak, bringing advanced security research to organizations of every size, independent researchers and open-source maintainers.