133/69 Monday, March 9, 2026
OpenAI officially launched Codex Security on Friday, introducing an AI-powered security agent designed to automatically identify, verify, and suggest fixes for vulnerabilities in source code. The feature is currently available as a Research Preview for users of ChatGPT Pro, Enterprise, Business, and Edu plans, with the first month offered free of charge. Codex Security builds upon an earlier project called Aardvark, which OpenAI introduced in a private beta in October 2025. The system combines advanced reasoning capabilities from OpenAI’s frontier models with automated validation mechanisms, reducing the rate of false positives by more than 50% compared to the early prototype.
During the 30-day beta testing period, Codex Security analyzed more than 1.2 million commits across external code repositories. The system identified 792 critical vulnerabilities and over 10,561 high-severity vulnerabilities across major open-source projects, including OpenSSH, GnuTLS, PHP, and Chromium. Several known CVEs were also detected, such as vulnerabilities in GnuPG (CVE-2026-24881, CVE-2026-24882), GnuTLS (CVE-2025-32988, CVE-2025-32989), and Gogs (CVE-2025-64175, CVE-2026-25242). The system operates through three primary stages: analyzing repository structures to build a threat model, identifying and prioritizing vulnerabilities within a sandbox environment, and generating remediation patches aligned with real-world system behavior.
The release of Codex Security comes shortly after Anthropic introduced Claude Code Security, highlighting the rapidly intensifying competition in the AI-powered application security market. Both companies aim to integrate AI more deeply into the Secure Development Lifecycle (SDLC). However, cybersecurity experts emphasize that while AI tools can significantly reduce false positives and accelerate vulnerability detection, human expert review remains essential, particularly before deploying AI-generated patches in critical production systems.
Source https://thehackernews.com/2026/03/openai-codex-security-scanned-12.html