OpenAI launches GPT-5.5-Cyber, 85.6% CyberGym score, for trusted defenders
OpenAI just launched GPT-5.5-Cyber, an AI model built to spot software vulnerabilities way faster and more accurately than before, scoring 85.6% on CyberGym, a clear jump from GPT-5.5's 81.8%.
Right now, it's only available to select "trusted defenders."
This comes as the US clamps down on Anthropic's AI models, giving OpenAI a chance to grab global attention in cybersecurity.
Patch the Planet aids open-source security
OpenAI also kicked off "Patch the Planet" with partners like Trail of Bits and HackerOne, offering free security help to open-source projects.
They say their AI can scan massive codebases and uncover hidden threats that humans might miss.
The Five Eyes intelligence alliance warns these new AI models could shake up cybersecurity fast, so keeping an eye on this tech is more important than ever.