Codex App August 10, 2026 Changelog

This release introduces two new security tools, Daybreak Blue and Daybreak Red, designed for approved cybersecurity professionals. It matters to "defenders"—the people tasked with finding and fixing security flaws—because it gives them specialized AI models to safely discover vulnerabilities and verify that their patches actually work.

What's new

• Daybreak Blue: A new defensive mode for everyday security work. It uses general AI models to help find vulnerabilities, review code for safety, analyze malware, and verify that software patches are correct.
• Daybreak Red: A new offensive mode for advanced testing. It uses specially trained AI models to safely "attack" a system (penetration testing or red teaming) to see if vulnerabilities can actually be exploited. This requires its own separate approval to use.
• Strict Access Controls: To use these tools, you have to apply for permission through a "Trusted Access" system. Your approval is tied exactly to your specific account, project, and the specific tool you are allowed to use.

Improvements

• Safer Testing Environments: The update emphasizes working in an "isolated environment" (a safe, digital sandbox separate from your main system) and using "least-privilege" permissions. This means the AI is only given the bare minimum access required to do its job, preventing accidental damage to real systems.
• Auto-Review for Safety: There is a new focus on an "Auto-review" step. If the AI tries to do something that reaches outside its safe sandbox, the system automatically checks it first to make sure it's safe to proceed.

Breaking changes

No breaking changes in this release.