OpenAI launches GPT-6 Cyber with Daybreak security framework

OpenAI introduced GPT-6 Cyber on November 5 as a cybersecurity model for finding and validating exploitable vulnerabilities within researcher-defined scope. The company says it improved 51% over GPT-5.6 Cyber on ExploitGym and stayed within authorized boundaries in 99.93% of cases without a deployment prompt. OpenAI also paired the model with the Daybreak Blue and Daybreak Red framework, plus safeguards including misalignment monitoring, allowlisting for Daybreak Red API access and private safety processing.