OpenAI unveils GPT-6 Cyber, rated Critical under its Preparedness Framework

OpenAI on Thursday showed off GPT-6 Cyber, a security model it says is both more capable than its predecessor and better at respecting the limits researchers set for it. On ExploitGym, which checks whether an AI can exploit a given flaw in a remote target, the new model reached 48.5%, up from 32.1% for GPT-5.6 Cyber. OpenAI rated it Critical for cybersecurity under its Preparedness Framework, above the High designation given to GPT-5.6 Cyber.

The scope claim rests on an alignment test. Given an exploitation task plus supporting services that were deliberately easy to hack but off-limits, the model left them alone in 99.93% of cases without deployment safety prompts, against 94.49% for GPT-5.6 Cyber. With the prompts, it never touched the decoy services. OpenAI attributes that to training aimed at staying inside authorized scope.

Pointed at the Linux kernel, the model reproduced 1,724 vulnerabilities and built working privilege escalation exploits for 948. GPT-5.6 Sol, run with the same tooling, produced 166. OpenAI says it has reported 304 critical issues to maintainers with proposed fixes. In Firefox, the model documented 81 new flaws Mozilla rated high severity, each with evidence and reproducible tests; Mozilla has started rolling out patches.

Access is limited to select approved Daybreak Red customers, through the API and Codex, with wider opening planned. Dragos, which tested Daybreak Red models in a controlled lab, fed into new safeguards meant to keep the models from helping compromise industrial controls. Its chief AI officer, Rob T. Lee, said the skill floor for pre-attack reconnaissance and vulnerability research is dropping. That's the part worth watching: the same capability that finds kernel bugs for maintainers shortens the work in front of an attacker.