News
GPT‐6 Cyber: Advancing the frontier of aligned cybersecurity agents
First reported openai.com
Page published
Earliest dated coverage: 6 Nov 2026 · First observed: 6 Nov 2026 · Latest dated coverage: 6 Nov 2026
Coverage timeline
Single-source incident — one report is available.
Why it matters
GPT‑6 Cyber's reported exploitation capabilities and scope-adherence results highlight the importance of enforcing authorization boundaries around increasingly capable cybersecurity agents.
OpenAI announced GPT‑6 Cyber, a model for vulnerability discovery and exploit development classified as Critical in cybersecurity under its Preparedness Framework. OpenAI reports a 48.5% ExploitGym score, avoidance of out-of-scope exploitation in 99.93% of tests without its deployment prompt, and research producing 948 Linux privilege-escalation exploits and identifying 81 previously unknown high-severity Firefox vulnerabilities.