OpenAI launches Daybreak security initiative with GPT-5.5-Cyber and Codex Security agent

OpenAI has launched Daybreak, an AI initiative designed to detect and patch security vulnerabilities before attackers can exploit them. The system combines multiple OpenAI models including the Codex Security agent and newly released GPT-5.5-Cyber models.

Daybreak uses the Codex Security AI agent, which launched in March, to create threat models based on an organization's codebase. According to OpenAI, the system focuses on possible attack paths, validates likely vulnerabilities, and automates detection of higher-risk security issues.

Security-specialized models

The initiative incorporates specialized cyber models including GPT-5.5 with Trusted Access for Cyber and GPT-5.5-Cyber, which began rolling out last week. OpenAI states that "Daybreak brings together the most capable OpenAI models, Codex, and our security partners."

Pricing, context window size, and detailed specifications for GPT-5.5-Cyber have not been disclosed.

Following Anthropic's approach

The launch comes one month after Anthropic announced Claude Mythos, a security-focused AI model the company claimed was too dangerous for public release. Anthropic shared Mythos privately as part of Project Glasswing, though unauthorized parties reportedly gained access to the model.

Unlike Claude Mythos, OpenAI has not indicated that Daybreak will be restricted from public or commercial use. The company says it is working with "industry and government partners" as it prepares to "deploy increasingly more cyber-capable models."

What this means

OpenAI's entry into security-focused AI puts it in direct competition with Anthropic's approach, but with a more open deployment strategy. The integration of the existing Codex Security agent suggests OpenAI is building on tools already in limited use rather than starting from scratch. The lack of disclosed pricing or technical specifications makes it unclear whether Daybreak will be accessible to smaller organizations or remain an enterprise-only offering. The mention of government partnerships suggests OpenAI may be positioning these models for use in critical infrastructure and national security applications.