GPT-6 Cyber

GPT-6 Cyber is a cybersecurity-specialized model from OpenAI, announced on November 5, 2026 as the successor to GPT-5.6 Cyber. It is built to find and validate exploitable vulnerabilities, develop exploits for authorized testing, and help produce candidate patches. OpenAI states it received dedicated training to stay within authorized scope, follow restrictions on its actions, and report results accurately.\n\nAccording to OpenAI, it scores 48.5% on ExploitGym (up from 32.1% for GPT-5.6 Cyber) and reached 40.3% recall on a new offensive web penetration testing evaluation, versus 32.0% for GPT-6 Astra. On the new RemoteExploitBench it completed 95% of tasks, while GPT-6 Astra with Daybreak Blue refused all of them. In an ExploitGym Honeypot alignment test, it avoided exploiting out-of-scope services in 99.93% of cases without deployment prompts and 100% with them. OpenAI's researchers used it to develop 948 Linux privilege escalation exploits and to identify 81 previously unknown high-severity Firefox vulnerabilities. Mozilla has released fixes for many of them. OpenAI classifies the model as Critical in cybersecurity under its Preparedness Framework.\n\nAccess is not general. It is offered through the Daybreak program (Daybreak Red for deeper exploit development and authorized security testing), and creators describe it as restricted to approved security researchers. Cisco, Calif and CodeRabbit are named as users. The benchmark figures are vendor-reported.