OpenAI's GPT-5.5-Cyber is first frontier AI cleared by a US government review board for autonomous vulnerability discovery

OpenAI shipped GPT-5.5-Cyber — scoring 85.6% on CyberGym — to all ChatGPT subscribers and expanded Daybreak to 28 partner governments, making it the first frontier AI cleared by a US government review board for autonomous vulnerability discovery and patch generation. That authorization is the structural signal: frontier AI has passed the formal threshold for unattended offensive security work inside allied government infrastructure, at a scale that makes Daybreak a new baseline for national-level cyber operations.

Source: openai.com ↗