OpenAI has unveiled Daybreak, a new cybersecurity initiative designed to identify and fix security vulnerabilities before malicious actors can exploit them. The system leverages GPT-5.5-Cyber along with Codex Security, which first launched in March, to build threat models based on an organization's existing code infrastructure. Daybreak then prioritizes potential attack vectors, validates which vulnerabilities are most likely to be exploited, and automates the detection process for the highest-risk security gaps.
The timing of the announcement is notable, coming roughly a month after rival AI company Anthropic revealed its own security-focused project. Anthropic's Claude Mythos was positioned as so potentially dangerous that the company initially restricted it to private access under what they called Project Glasswing. Despite these precautions, at least some unauthorized individuals managed to obtain access to the model, raising questions about the feasibility of keeping highly capable security tools contained.
Unlike single-model solutions, Daybreak pulls together multiple specialized components. In addition to Codex Security and GPT-5.5-Cyber, which began rolling out last week, the system incorporates GPT-5.5 with Trusted Access for Cyber capabilities. OpenAI has emphasized that the initiative involves collaboration with both industry partners and government agencies as the company works toward deploying increasingly sophisticated cyber-capable models.