OPENAI · 2026-11-05 18:27 UTC

GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents

OpenAI introduced GPT-6 Cyber, a specialized cybersecurity model showing 51% relative improvement over GPT-5.6 Cyber on ExploitGym benchmark, scoring 48.5% vs 32.1%. The model demonstrates strong alignment, avoiding out-of-scope exploitation in 99.93% of tests and has already helped identify 81 previously unknown vulnerabilities in Firefox and 948 privilege escalation exploits in Linux. OpenAI plans to expand Daybreak program to bring GPT-6 Cyber capabilities to organizations of all sizes, independent researchers, and open-source maintainers for authorized security research.

Loading the story on Tree News…