OpenAI Daybreak uses GPT-5.5 to find software vulnerabilities
OpenAI today unveiled Daybreak, a response to Anthropic’s Project Glasswing and the Mythos AI model. Like Glasswing, Daybreak is a cyber defense project designed to help technology companies detect security vulnerabilities in their platforms.
OpenAI explains that Daybreak aims to integrate cybersecurity into software from the very beginning. It builds on OpenAI’s April release of GPT-5.4-Cyber, which the company says has helped to fix over 3,000 security vulnerabilities.
Daybreak combines the intelligence of OpenAI models, the extensibility of Codex as an agent-based platform, and the expertise of our partners across the entire security chain to make the world a safer place for everyone. Security professionals can integrate secure code reviews, threat modeling, patch validation, dependency risk analysis, detection, and remediation guidance into the daily development process, making software more resilient from the start.
OpenAI CEO Sam Altman stated that OpenAI aims to collaborate with as many companies as possible to continuously protect their software from cyber threats. Several companies, including Apple, Microsoft, Google, and Amazon, have already implemented Anthropic’s competing Glasswing program.
Daybreak creates customizable threat model
Daybreak uses Codex Security to build a customizable threat model from an organization’s software repository and then automate monitoring for critical vulnerabilities. Identified issues can be investigated in an isolated environment.
Companies can request a Daybreak assessment from OpenAI, which also includes a vulnerability scan. Pricing is not listed.
Three models are available. GPT-5.5 offers standard security measures for general use, while GPT-5.5 with Trusted Access for Cyber is designed for verified defense tasks in authorized environments. GPT-5.5-Cyber is intended for specialized, authorized workflows and features stronger verification and account-level controls.
OpenAI is working with partners from industry and government to deploy more cyber-enabled models in the future.
