OpenAI has released GPT-6 Cyber, a specialized AI model designed to identify and correct security vulnerabilities. The model scored 48.5% on ExploitGym, a benchmark that measures the ability to exploit specific vulnerabilities—a significant jump from GPT-5.6 Cyber's 32.1% performance.
Beyond raw detection capability, GPT-6 Cyber demonstrates strong security restraint. The model avoided exploiting out-of-scope services in 99.9% of cases, even without additional safety instructions built in. This built-in caution suggests OpenAI prioritized responsible behavior alongside technical capability.
Access to GPT-6 Cyber is currently restricted to a specialized group of cybersecurity experts, limiting its availability to organizations and researchers who can use it safely. This controlled rollout reflects the sensitivity of deploying AI tools with exploitation capabilities.