Back
AI Models5 November 20261 min read

OpenAI releases GPT-6 Cyber for authorized security research

OpenAI has launched GPT-6 Cyber, a specialized model trained to find and validate software vulnerabilities while staying within the boundaries set by researchers.

OpenAI releases GPT-6 Cyber for authorized security research

About the news

The model carries a “Critical” cybersecurity rating under OpenAI’s Preparedness Framework, the highest tier.

Performance gains

On the ExploitGym benchmark, which measures a model’s ability to exploit a specified vulnerability, GPT-6 Cyber scores 48.5%, up from the 32.1% achieved by the previous cyber-focused model (GPT-5.6 Cyber). In internal research, the model helped develop working exploits for 948 Linux kernel privilege-escalation issues and identified 81 previously unknown high-severity vulnerabilities in Firefox. Mozilla has begun releasing fixes for many of the Firefox findings.

Stronger alignment

OpenAI emphasizes improved reliability in staying within authorized scope. In alignment evaluations, GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of tests without its deployment prompt, and 100% of the time when the prompt was included.

Access and purpose

GPT-6 Cyber is available only to selected customers in OpenAI’s Daybreak Red program, which supports authorized defensive security work such as vulnerability research, exploit development, and penetration testing. It is not broadly available. The release illustrates how frontier models are being applied to real defensive cybersecurity, with tighter controls on scope and a focus on helping security teams find and fix weaknesses before they can be misused.

Sources