OPENAI BUILT A CYBERWAR AI AND WON’T LET JUST ANYONE NEAR IT OpenAI launched the full version of GPT-5.5-Cyber on June 22 and locked it behind a restricted access program that keeps it out of the public API entirely. This is not a product you can sign up for. Organizations must go through OpenAI’s Trusted Access for Cyber program to get near it, and even then access is gated to verified defensive security operations. The reason for the caution is visible in the benchmarks. GPT-5.5-Cyber scored 85.6 percent on CyberGym, compared to 81.8 for standard GPT-5.5. On ExploitGym, which tests how well a model can identify and exploit real vulnerabilities, it hit 39.5 percent against 25.95 for the base model. On SEC-bench Pro, 69.8 percent versus 63.1. OpenAI is calling this the highest CyberGym score any single model has ever recorded. The model can autonomously scan large codebases, identify security-critical components, validate suspected vulnerabilities, write patches, and test them. OpenAI is also launching a Daybreak Cyber Partner Program, letting vetted security vendors embed the model into commercial products. What is notable is what OpenAI is not doing. It is not open-sourcing this. It is not making it available on demand. It is building walls around a model it knows could be turned toward offensive use as easily as defensive. For now, at least, the walls are holding. Keywords: OpenAI GPT-5.5-Cyber, AI cybersecurity model, Daybreak cyber program, AI security