OpenAI Introduces GPT-6 Cyber for Approved Security Teams, Enhancing Vulnerability Research
OpenAI has launched GPT-6 Cyber, a specialized version of its GPT-6 Astra model, specifically trained for authorized vulnerability research. This new model is accessible to a select group of approved customers within OpenAI's Daybreak Red program, which focuses on offensive-style cybersecurity work like penetration testing and red teaming. OpenAI describes GPT-6 Cyber as a significant advancement in security research. The model has demonstrated impressive capabilities, including identifying 81 previously unknown high-severity vulnerabilities in Firefox, for which Mozilla has already shipped fixes. In separate research on the Linux kernel, GPT-6 Cyber reproduced 1,724 vulnerabilities and developed working exploits for 948 of them, leading to OpenAI disclosing 304 critical findings and candidate patches to Linux maintainers. On the ExploitGym benchmark, GPT-6 Cyber scored 48.5%, an improvement from GPT-5.6-Cyber's 32.1%.
