OpenAI Launches GPT-6 Cyber With Restricted Access and Scope Controls
Reported by all-ai, OpenAI and Unite.AI3 sources ↓
Latest in this story: OpenAI launches GPT-6 Cyber with 'Critical' cybersecurity rating
OpenAI introduced GPT-6 Cyber on November 5, 2026, with API and Codex access for a select group of approved Daybreak Red customers. The model is intended for authorized vulnerability research and exploit development, with safeguards to limit activity to approved scope.
Key points
- Access is initially limited to approved Daybreak Red customers, with a gradual expansion planned
- OpenAI reports a 48.5% ExploitGym score, compared with 32.1% for GPT-5.6 Cyber
- Mozilla classified 81 previously unknown Firefox vulnerabilities identified by the model as high severity
- OpenAI classifies GPT-6 Cyber as Critical for cybersecurity, up from High for GPT-5.6 Cyber
- OpenAI’s system card reports a 0.07% out-of-scope honeypot exploitation rate without deployment safety instructions
Sources
Details
GPT-6 Cyber is a cybersecurity variant of GPT-6 Astra, with API and Codex access initially restricted to approved Daybreak Red customers. Unite.AI reports that applicants undergo identity and eligibility checks, and that OpenAI plans to expand access across the program. GPT-5.6 Cyber is also becoming available to Daybreak Red customers.
OpenAI reports a 48.5% score on ExploitGym, which tests whether a model can exploit a specified vulnerability; GPT-5.6 Cyber scored 32.1% and GPT-6 Astra 42.4%. On an initial RemoteExploitBench evaluation involving known vulnerabilities in GitLab, PostgreSQL and Jenkins, GPT-6 Cyber completed 95% of tasks, according to OpenAI. In Irregular’s external FrontierCyber evaluation, it solved 90 of 288 challenges, fewer than Astra’s 96; neither model successfully attacked fully hardened targets or solved the 15 hardest challenges.
OpenAI says the model provided evidence and reproducible tests for 81 previously unknown Firefox vulnerabilities that Mozilla classified as high severity, and identified 30 known issues as more severe than previously reported. Its Linux research report describes reproducing 1,724 vulnerabilities and producing 948 local privilege-escalation exploits; human researchers reviewed the results before reporting them to maintainers. OpenAI says the work used $1,213,430 in API credits.
OpenAI classifies GPT-6 Cyber as Critical for cybersecurity. Its system card reports that the model exploited an out-of-scope auxiliary honeypot in 0.07% of graded tests without deployment safety instructions and in none with them. Deployment measures include monitoring, default API IP allowlisting and controls that can pause or terminate requests; the external evaluation results also show limits in performance against difficult or hardened targets.
Written from coverage as of
Images
- all-ai
- Unite.AI
- all-ai
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh