GPT-6 Cyber Receives Critical Rating and Is Available Only to Verified Specialists
OpenAI has introduced GPT-6 Cyber, a model for identifying vulnerabilities and preparing exploits. For now, only verified specialists with authorization are permitted to test it.
In OpenAI's classification, the development has been assigned the Critical level. This signifies a higher potential cyber risk than the High rating of GPT-5.6 Cyber.
With the model's participation, researchers discovered 81 previously unknown vulnerabilities in Firefox, the company reported. Mozilla deemed them serious and began releasing updates. In the Linux kernel, GPT-6 Cyber helped prepare working exploits for 948 vulnerabilities. According to OpenAI, 588 of these had not been previously recorded. Data on 304 critical issues, along with fix options, were sent to developers.
Due to the model's capabilities, its distribution was decided to be restricted. At this stage, GPT-6 Cyber is available only to verified members of Daybreak Red conducting authorized security assessments. OpenAI plans to grant access to more specialists within the coming weeks.
