OpenAI has launched GPT-6 Cyber, a cybersecurity-focused model designed to find, validate and help fix software vulnerabilities while staying within researchers’ authorized scope.
The company says the model is initially available through its API and Codex to a select group of approved Daybreak Red customers, with access expanding to all approved customers in the coming weeks.
OpenAI rates GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber. In the company’s evaluations, the model scored 48.5% on ExploitGym, compared with 32.1% for GPT-5.6 Cyber. It avoided exploiting out-of-scope services in 99.93% of tests without deployment safety prompts and 100% with those prompts. OpenAI says the model’s safeguards are designed to keep testing within the boundaries set by researchers, a concern also raised in Apple’s warning about AI agents acting without authorization.
OpenAI said GPT-6 Cyber helped identify 81 previously unknown high-severity Firefox vulnerabilities and develop working exploits for 948 Linux kernel vulnerabilities after reproducing 1,724. The company disclosed 304 critical Linux findings to maintainers with candidate patches. Mozilla has begun releasing fixes for many of the Firefox issues in its Firefox 159 security advisory. The Linux findings are part of the same research pipeline detailed in OpenAI’s November 4 report, which Abijita covered in its report on 948 Linux privilege-escalation flaws.
The new safeguards include cyber-specific monitoring for activity outside a user’s authorization, additional checks for high-risk requests, and IP allowlisting for Daybreak Red API access by default. OpenAI says the model is intended for authorized security research and that it has added protections against misuse involving critical infrastructure. GPT-5.6 Cyber is also becoming available to all Daybreak Red customers.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
