OpenAI Releases Specialized Cybersecurity Model GPT-6 Cyber
OpenAI 发布 GPT-6 Cyber，专攻网络安全
OpenAI has officially launched GPT-6 Cyber, a specialized cybersecurity model based on GPT-6 Astra, elevating its cybersecurity capability tier to "Critical" under its Preparedness Framework. The model identified 81 previously unknown high-severity vulnerabilities in Firefox, and reproduced 1,724 vulnerabilities while crafting 948 privilege escalation exploits in Linux kernel testing. It also scored 48.5% on ExploitGym, with its honeypot breach rate dropping to 0.07% in the absence of safety prompts.
Drawing on benchmark data from multiple real-world codebases and industrial scenarios, the report highlights the quantum leap specialized cybersecurity large models have achieved in vulnerability discovery and exploitation.
Source:AITNT · AI头条（网页） · aitntnews.com