OpenAI names GPT-6 Cyber: Critical cyber rating, 81 novel Firefox highs, Daybreak Red gate — Linux 948-LPE pipeline was this model
- What changed vs the Nov 4 note: the unnamed 'upcoming cyber model' is now GPT-6 Cyber; Firefox campaign numbers and the Critical rating are new; Daybreak Red / Blue split and critical-infrastructure defender track are spelled out.
- Alignment claim to weigh: ExploitGym Honeypot (inspired by the Hugging Face incident) — 99.93% avoided out-of-scope services without the deployment prompt, 100% with it (vs 94.49% for GPT-5.6 Cyber).
- Ops implication: expect faster patch volume from vendors using these agents; treat unauthenticated 'AI found N kernel bugs' social posts as marketing until a CVE or maintainer ack lands.
- Related on-site:
openai-thousand-fewer-paths-to-root-948-linux-lpe-pipeline(Nov 4).
Sources: OpenAI: Introducing GPT-6 Cyber (Nov 5) · OpenAI: A thousand fewer paths to root (Nov 4) · BleepingComputer: OpenAI's new GPT-6 Cyber AI finds 81 Firefox flaws, 948 Linux bugs