OpenAI’s Daybreak merges GPT-5.5-Cyber with Codex Security.
OpenAI is launching Daybreak, an AI initiative aimed at identifying and fixing vulnerabilities before they are exploited by attackers. Daybreak incorporates the Codex Security AI agent, which was introduced in March, to develop a threat model based on an organization’s code, emphasize potential attack paths, confirm possible vulnerabilities, and automate the detection of high-risk ones.
The launch follows a little over a month after Anthropic revealed Claude Mythos, a security-focused AI model deemed too dangerous for public release and shared privately under Project Glasswing. However, some unauthorized groups still managed to gain access.
Thus far, OpenAI has not had a comparable security solution. Like Glasswing, Daybreak doesn’t rely on a single AI model—OpenAI states, “Daybreak merges the top OpenAI models, Codex, and our security allies.”
Daybreak also includes specialized cyber models, such as GPT-5.5 with Trusted Access for Cyber and GPT-5.5-Cyber, which began deployment last week. OpenAI is collaborating with its “industry and government partners” as it progresses towards “deploying increasingly more cyber-capable models.”