OpenAI Releases GPT-6 Cyber for Enhanced Cybersecurity

OpenAI has launched GPT-6 Cyber, a new cybersecurity model designed to find and validate exploitable vulnerabilities within set boundaries. This model is built to enhance the identification and fixing of vulnerabilities, demonstrating significant improvements over its predecessor, GPT-5.6 Cyber, with a 51% relative improvement on the ExploitGym benchmark. It is available for deployment in real-world software environments, including Linux and Firefox codebases, where it has already identified numerous exploits and vulnerabilities. According to OpenAI, GPT-6 Cyber is more reliable in adhering to task boundaries, avoiding out-of-scope services in 99.93% of tests without deployment prompts and 100% with them. This model is classified as Critical in cybersecurity under OpenAI’s Preparedness Framework, marking a step up from the High classification of GPT-5.6 Cyber.