OpenAI introduced Codex Security, an AI-powered security agent designed to automatically identify, validate, and propose fixes for vulnerabilities in software codebases. Released in a research preview to ChatGPT Pro, Enterprise, Business, and Edu users via the Codex web interface, the tool builds deep contextual understanding of a project to uncover complex security issues that traditional automated scanners may miss while reducing noise from insignificant findings. Codex Security evolved from OpenAI’s earlier Aardvark project introduced in private beta in October 2025 and combines advanced model reasoning with automated validation workflows. During the past month of beta testing, the system analyzed over 1.2 million commits across external repositories, identifying 792 critical and 10,561 high-severity vulnerabilities in widely used open-source projects such as OpenSSH, GnuTLS, GOGS, Thorium, libssh, PHP, and Chromium, including issues tracked as CVE-2026-24881, CVE-2026-24882, and CVE-2025-32988. The platform operates through a three-stage process. First, it analyzes repository structure to generate a security-focused threat model, then detects and classifies vulnerabilities based on real-world impact, and finally validating them in a sandboxed environment to reduce false positives and potentially generate proof-of-concept exploits. Once verified, the agent proposes fixes aligned with system behavior to minimize regressions and streamline remediation. OpenAI reports that repeated scans have improved precision significantly, cutting false positives by more than half, positioning Codex Security as part of the emerging class of AI-driven application security tools alongside offerings like Anthropic’s Claude Code Security. CTIX analysts will continue to report on novel vulnerabilities and the ways in which they can be defended against.