OpenAI launches GPT-6 Cyber for vetted security researchers

OpenAI released GPT-6 Cyber, a cyber-specialized variant of GPT-6 Astra, to select approved Daybreak Red customers through the API and Codex. The company said it plans to expand access to all Daybreak Red customers “soon,” but did not set a date or disclose pricing.

The model is designed for authorized vulnerability discovery, exploit development, web penetration testing and software remediation. OpenAI classified GPT-6 Cyber as “Critical” in cybersecurity under its Preparedness Framework, a capability designation made by the company rather than an independently assigned rating.

In OpenAI’s evaluation, GPT-6 Cyber scored 48.5% on ExploitGym, which tests whether a model can exploit a specified vulnerability, compared with 32.1% for GPT-5.6 Cyber. It also said the model exploited an out-of-scope honeypot in 0.07% of graded rollouts without deployment safety instructions and in 0% with those instructions.

Access is restricted through Daybreak Red. OpenAI’s system card says organizations must use controls including single sign-on, multi-factor authentication, managed devices and least-privilege access, while individual applicants face identity, liveness and professional-suitability checks. Daybreak Red API access uses IP allowlisting by default. The company also says it uses cyber-specific and account-compromise monitoring, blocks high-risk assistance involving critical systems, and recommends sandboxing, action review and human oversight for high-risk workflows.

The system card says external evaluator Irregular observed no successful attacks on fully hardened targets. It also says neither GPT-6 Cyber nor GPT-6 Astra solved any of the 15 highest-difficulty FrontierCyber challenges.

OpenAI said GPT-6 Cyber produced evidence and reproducible tests for 81 previously unknown Firefox vulnerabilities that Mozilla classified as high severity, and helped propose patches. Mozilla’s Firefox 159 advisory lists many high-impact fixes credited to OpenAI researchers. It does not independently attribute the complete set of 81 to GPT-6 Cyber.

The launch follows OpenAI’s earlier disclosure of a Linux kernel vulnerability pipeline. In research on that pipeline, OpenAI reported reproducing 1,724 vulnerabilities and developing working exploits for 948 local-privilege-escalation vulnerabilities in a controlled environment. The pipeline generated candidate patches before human researchers reviewed its outputs.