OpenAI Launches Codex Security: AI Agent for Automated Vulnerability Discovery and Remediation

OpenAI rolled out Codex Security on March 6, 2026, an AI-powered application security agent evolved from its internal research project Aardvark that can autonomously find, validate, and propose fixes for software vulnerabilities. The agent represents a significant step beyond code completion tools — operating end-to-end across the vulnerability lifecycle from initial discovery through remediation, without requiring a human to initiate each scan. The launch arrives in the same week that Mozilla reported Claude Opus 4.6 found 100+ Firefox bugs in two weeks, signaling a broader shift toward AI-assisted security engineering as a production practice.

Key Takeaways

• OpenAI Codex Security can autonomously discover, validate, and propose fixes for vulnerabilities in application code — evolved from OpenAI's internal Aardvark research project, launched March 6, 2026
• Agent operates across the full vulnerability lifecycle without per-scan human prompts — designed for integration into CI/CD pipelines and security review workflows
• Covered by Axios (Sam Sabin) and Techmeme; arrives same week as Mozilla's Claude AI red team finding 22 CVEs in Firefox, establishing AI-native security tooling as a practical category

Original source: Axios / OpenAI