OpenAI has begun giving its GPT-6 Cyber model to a limited group of approved security teams, saying it can find and test software weaknesses and help developers prepare fixes, including for flaws Mozilla is already fixing.
Announced on November 5, the cybersecurity model is initially available through OpenAI’s Daybreak Red program to selected customers doing approved security work, rather than to the general public. OpenAI says its researchers used the model to find 81 previously unknown, high-severity Firefox security weaknesses and help develop possible fixes. Mozilla has released fixes for many of those weaknesses.
The company also says GPT-6 Cyber developed 948 methods that use Linux security weaknesses to gain higher access rights. It reported Linux security findings to system maintainers along with proposed fixes. OpenAI rated the model “Critical” for cybersecurity under its safety framework, one level above the “High” rating given to its predecessor, GPT-5.6 Cyber.
The company says that in one evaluation, the model did not go beyond approved testing limits in any case when safety instructions for deployment were enabled. However, the figures and research results come from OpenAI’s own reports. The restricted rollout reflects the risk that the same ability to find and demonstrate security weaknesses could help defenders or attackers if misused.