What happened
OpenAI published “GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents” on November 5, 2026. On ExploitGym, GPT-6 Cyber scores 48.5% against 32.1% for GPT-5.6 Cyber — a 51% relative improvement on OpenAI’s account. In an alignment evaluation (ExploitGym Honeypot), it avoided out-of-scope services in 99.93% of tests without the deployment prompt and 100% with it, up from 94.49% for GPT-5.6 Cyber without the prompt. OpenAI classifies the model as Critical for cybersecurity under its Preparedness Framework, up from High. The post ties the model to the earlier Linux campaign (948 working LPEs) and to 81 previously unknown high-severity Firefox vulnerabilities with Mozilla shipping fixes. Access is initially through the API and Codex for a select group of Daybreak Red customers, with GPT-5.6 Cyber opened to all Daybreak Red customers.
Why it matters
Readers already have the 948 Linux count without a model name. This post names GPT-6 Cyber, puts a public-style ExploitGym score next to it, and states the access door.
RemoteExploitBench: OpenAI says GPT-6 Cyber completed 95% of tasks while Astra with Daybreak Blue refused 100%.
Web penetration recall without source: 40.3% for GPT-6 Cyber vs 32.0% for GPT-6 Astra.
Daybreak Blue remains the defender path with fewer refusals; Daybreak Red is the deeper exploit path.