On March 6, 2026, OpenAI introduced Codex Security, a novel AI agent engineered to enhance code security reviews. This tool aims to pinpoint complex vulnerabilities that traditional security systems frequently miss, offering actionable solutions to bolster overall system integrity and streamline secure code deployment.
As the pace of software development accelerates, especially with AI's influence, security challenges have intensified. Unlike conventional AI security tools that often produce numerous low-impact alerts, Codex Security reduces the burden on human teams by accurately identifying real threats. It builds on insights from OpenAI’s earlier agent, Aardvark, which was launched in October 2025 and showed improved detection accuracy by decreasing false positives.
Codex Security begins by assessing a project's repository to create a customized threat model that outlines its functions and vulnerabilities. This model aids in prioritizing threats based on their potential impact, ensuring thorough investigations via sandbox testing. Additionally, the system suggests fixes that harmonize with existing functionality, allowing developers to focus on critical security issues effectively.