OPENAI’S NEW CYBERSECURITY MODEL JUST SET THE HIGHEST BENCHMARK SCORE EVER RECORDED FOR VULNERABILITY DETECTION OpenAI is now in the business of hunting security vulnerabilities, and its new model is better at it than anything that has come before. GPT-5.5-Cyber launched as the centerpiece of OpenAI’s expanded Daybreak cybersecurity initiative. The number that matters most: 85.6 percent on CyberGym, the industry benchmark for AI cybersecurity capability. The previous version of the model scored 81.8 percent. Every competing system that OpenAI is aware of scores lower than the new mark. The model does more than identify vulnerabilities. It can trace whether vulnerable code is actually reachable in production, validate findings in controlled environments, generate and test patches, and prepare detailed evidence packages for human review. The whole pipeline from detection to remediation, handled by a single model, operating at a scale no human security team can match. OpenAI is rolling this out carefully. Access is restricted to verified defenders through a limited release with extra monitoring. The company has also expanded Codex Security for coding teams and launched Patch the Planet, a partnership with Trail of Bits that brings AI-assisted security to open-source software maintainers who typically have no security budget at all. The dual-use concern is obvious. A model that can find and exploit vulnerabilities faster than any previous system is extremely valuable to attackers as well as defenders. OpenAI says it has built in controls to prevent offensive use, but the gap between a powerful security AI and a powerful attack AI is narrower than most people realize.