OpenAI has introduced Daybreak, a new suite of AI-driven agents designed to automate critical phases of the software security lifecycle, including penetration testing and vulnerability remediation. The platform is divided into two primary components: Red Team, which focuses on identifying and validating security flaws, and Defense Factory, which assists in the development and implementation of code fixes. Currently available through an early-access program, the suite aims to bridge the gap between discovering a vulnerability and successfully deploying a patch.
The Red Team component functions as an automated penetration testing tool that analyzes application source code and operational context to map potential attack vectors. According to OpenAI, this system is capable of verifying findings generated by traditional static code scanners, bug bounty programs, and vulnerability disclosure reports. By confirming the legitimacy of these alerts, the tool provides security teams with the necessary evidence to prioritize remediation efforts. Furthermore, Red Team can be used to test the efficacy of a patch once it has been developed, ensuring that the proposed fix successfully blocks the identified exploit.
To maintain operational security, OpenAI has implemented strict boundaries for the Red Team agents. These assessments are conducted within isolated environments, and the system utilizes a feature called Guardian review to restrict the specific applications and endpoints that the agents are permitted to access. This design is intended to prevent unauthorized lateral movement or unintended interactions during the testing phase.
The second component, Defense Factory, is designed to streamline the remediation process. This tool integrates with existing security infrastructure to investigate flagged vulnerabilities, trace the root cause through the codebase, and generate proposed code fixes. The system is currently available as a plugin for ChatGPT, allowing users to connect directly to their GitHub repositories. Once connected, the agents can analyze the default branch of a repository or focus on specific folders and files.
Defense Factory offers several deployment options for development teams, including the ability to monitor new commits on a default branch or schedule recurring automated scans. It also supports security reviews for pull requests, providing an additional layer of oversight before code changes are merged into the production environment. For developers working locally, the platform includes a desktop application that can scan local checkouts and uncommitted changes. This desktop version features a Deep scan mode, which performs multiple passes over the code to increase the thoroughness of the analysis, albeit at the cost of additional processing time and token usage.
While the agents automate the investigation and drafting of patches, OpenAI emphasizes that human oversight remains a core requirement of the workflow. Once the Defense Factory generates a proposed fix, engineers are expected to review and test the changes before they are finalized. The platform includes a feature called Codex that allows teams to examine the evidence gathered by the agents, provide additional context, and manually refine the proposed modifications.
The integration requires a personal GitHub connection for any individual initiating a scan, and all cloud-based analysis is processed within the Codex cloud environment. By combining the discovery capabilities of Red Team with the remediation workflows of Defense Factory, the platform seeks to reduce the time between the identification of a security weakness and the deployment of a verified solution. OpenAI has positioned these tools as a way to leverage existing security investments by automating the more labor-intensive aspects of vulnerability management.





