OpenAI announced GPT-6 Cyber Thursday and classified it as Critical for cybersecurity, a designation accompanying restricted access for security customers as the model is designed to develop software exploits.
OpenAI describes it as a GPT-6 Astra variant intended for authorized security assessments, including finding flaws, evaluating exploits and testing web applications. The system is initially available through the API and Codex to selected, approved Daybreak Red customers.
OpenAI says its researchers reproduced 1,724 Linux kernel vulnerabilities and developed 948 working exploits. The company said it disclosed 304 critical findings to maintainers with candidate patches.
OpenAI says its researchers generated supporting evidence and repeatable tests for 81 previously unknown Firefox flaws that Mozilla rated high severity. It also said the team identified 30 known issues it judged more serious than earlier assessments, after which Mozilla began applying fixes. Mozilla’s Oct. 27 advisory for Firefox 159 lists high-impact vulnerabilities and credits OpenAI reporters on some entries, but does not give an aggregate total of 81.
Brian Grinstead, Distinguished Engineer, Firefox, said, “Firefox entered 2026 as one of the world’s most hardened and scrutinized codebases.”
On OpenAI’s ExploitGym evaluation, GPT-6 Cyber scored 48.5%, compared with 32.1% for GPT-5.6 Cyber. OpenAI’s task-control review showed boundary compliance in 99.93% of tests without the deployment prompt; the figure reached 100% when the prompt was included.
The company says it monitors for activity beyond a user’s authorization and enables IP allowlisting by default for Daybreak Red API access. OpenAI’s system card says access requires identity verification and organizational safeguards, including single sign-on and multifactor authentication.
OpenAI has previously described a separate Daybreak Cyber Partner Program for approved security providers building commercial services with its models, covered in EconoTimes’ earlier reporting on OpenAI’s cybersecurity business. In a Nov. 4 report on Linux research, OpenAI said its workflow checks whether vulnerabilities can be reproduced, assesses exploitability, prepares candidate fixes and includes human review.
OpenAI plans to extend GPT-6 Cyber access to all Daybreak Red customers but has not announced when. GPT-5.6 Cyber is being made available to the full group.
