News
A thousand fewer paths to root
First reported openai.com
Page published
Earliest dated coverage: 5 Nov 2026 · First observed: 5 Nov 2026 · Latest dated coverage: 5 Nov 2026
Coverage timeline
Single-source incident — one report is available.
Why it matters
GPT-6 Cyber's reported residual scope violations highlight why autonomous exploitation agents need independently enforced authorization boundaries rather than model-level compliance alone.
OpenAI announced GPT-6 Cyber, claiming improved vulnerability discovery and exploitation alongside stronger compliance with authorized testing boundaries, although 0.07% of evaluation cases reportedly exceeded scope. A related OpenAI research report describes a prototype autonomous pipeline that generated 948 working Linux local privilege-escalation exploits and candidate patches.