OpenAI launches GPT-6 Cyber, cutting out-of-scope exploitation to 0.07%
gdb · x · 2026-11-06
OpenAI cofounder Greg Brockman announced GPT-6 Cyber, pitched as a major advance for security research:
- Already helping defend security-critical software including the Linux kernel and Firefox; security teams at Mozilla, Palo Alto Networks, Calif, and Cisco are using it
- Heavy investment in trust: the model is trained to stay within operator-specified scope. On the ExploitGym Honeypot eval (out-of-scope exploitation rate, lower is better, without deployment safety prompts), GPT-6 Cyber hits 0.07% vs 5.51% for GPT-5.6 Cyber
- Also launching Daybreak Red for Critical Infrastructure, letting approved defenders run authorized testing of critical systems
Access is limited to select Daybreak Red customers and will expand over time.
Related event: OpenAI Launches GPT-6 Cyber, a Security-Research-Tuned Model(9 posts)→
More from Models
- webAI Claims 12B Model Beats Gemini and Claude, But the Benchmark Claims Raise Eyebrows — eurie_kim · 2026-11-06
- User cancels OpenAI over Codex quota change, says Claude did a week's work in 4.5 hours — LethalLuck7 · 2026-11-06
- Claude actually pulls off a Rickroll when a user dares it to try — emollick · 2026-11-06
- NVIDIA touts time-to-market and quality benefits of its seven open AI models — karlfreund · 2026-11-06
- China Telecom's Xing4.0-29B-A4B lands llama.cpp support: trained fully on Ascend NPUs — jacek2023 · 2026-11-06
- Ripple-65B on one RTX PRO 6000: 110-120 tok/s on code in local agent tests — DZhang50 · 2026-11-06