A new OpenAI model called GPT-6 Cyber found 948 privilege-escalation vulnerabilities in the Linux kernel and built working exploits for most of them, the company said on November 5.
OpenAI said the model combed through roughly 2.8 million kernel-code lines, about 7% of the project, and produced 1,169 working exploits. The search cost $1.2 million ($1,213,430 in API credits), or about $1,279 per vulnerability.
That is five times more than an earlier model, GPT-5.6 Sol, found in a similar test last July, when it uncovered 166 vulnerabilities for $513,690. The newer model also found 81 previously unknown vulnerabilities in Firefox, according to OpenAI.
Max Justicz, chief technology officer and co-founder of Charge Robotics, independently confirmed the model's effectiveness after spending $5,000 on subscriptions to find a critical Ubuntu bug tracked as CVE-2026-109518. Justicz said: “I literally just aimed ChatGPT at the same body of code and said ‘I found this bug previously...’”
The Debian Project issued a patch, and no exploitation has been reported.
OpenAI also launched Daybreak, a $1 billion program aimed at cybersecurity professionals who protect critical infrastructure, which the company said already includes thousands of experts.
