OpenAI unveils GPT-5.5-Cyber, a security model built to find and fix flaws

OpenAI announced on Monday night the full release of GPT-5.5-Cyber, a model aimed at trusted cyber defenders. The company says the new system can analyze large codebases and repair security vulnerabilities quickly, with the goal of making software patching accessible at machine speed and reducing the need for slow, manual specialist work.

According to OpenAI, cyber defense has shifted from finding bugs to coping with a flood of alerts and delayed fixes. The new model is designed to automatically generate patches for browsers, network infrastructure and complex operating systems such as Linux, so attackers cannot exploit weaknesses before they are closed.

OpenAI also published performance numbers showing it ahead of Anthropic’s Mythos 5. In the CyberGym benchmark, which measures vulnerability reproduction in software environments, GPT-5.5-Cyber scored 85.6 percent, compared with 83.8 percent for Mythos 5. OpenAI said this is the highest score ever recorded for a single model in the category. The model also showed stronger “liberation” capabilities, meaning it can carry out complex cyber actions without unnecessary safety restrictions, build hypothetical attack scenarios to verify suspicions, and prepare evidence for human security teams.

The company tied the launch to Israel’s cyber sector, saying Daybreak partners include Wiz, Check Point and Cato Networks, which will be able to integrate the model into their security products to help protect critical infrastructure worldwide. OpenAI also said it is expanding cooperation with the governments of the United States, Britain, the European Union and Japan on customized defenses for sensitive government networks. The company said its goal is “a world of safer software and cyber resilience.”