OpenAI launches GPT-6 Cyber after it found 81 Firefox flaws
The specialized security model improves exploit success by 51% and is initially limited to approved Daybreak Red customers.
OpenAI has launched GPT-6 Cyber, a specialized model for authorized vulnerability research and exploit development. It is initially available through the API and Codex to a select group of approved Daybreak Red customers, with wider access planned.
The company says the model scored 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber — a 51% relative improvement. It also outperformed GPT-6 Astra on most cyber tasks and is classified Critical for cybersecurity under OpenAI’s Preparedness Framework.
GPT-6 Cyber is already behind several real-world security campaigns. OpenAI says it found 81 previously unknown high-severity Firefox vulnerabilities, helped reproduce 1,724 Linux kernel flaws and produced working exploits for 948 of them. Mozilla has begun shipping fixes for many of the Firefox issues. The model also helped validate a critical remote-code-execution path involving Brotli.
OpenAI says alignment training made the model more reliable at staying inside an authorized target scope. In one evaluation, it avoided exploiting out-of-scope services in 99.93% of tests without a deployment safety prompt and 100% with one.
This is not a general ChatGPT release. Daybreak Red access requires approval, IP allowlisting is enabled by default, and OpenAI recommends sandboxing and human review for high-risk workflows.
OpenAI
