OpenAI’s Codex Security Built to Automate Vulnerability Discovery and Remediation

OpenAI announced Codex Security (formerly Aardvark), a research-preview AI application security agent that analyzes repository context to generate project-specific threat models, automatically find and validate vulnerabilities in sandboxed environments, and propose tailored fixes. The announcement highlights substantial reductions in alert noise and false positives during beta testing, notes that the tool identified hundreds of high and critical findings (including assigned CVEs such as CVE-2025-35430) across open-source projects, and describes a "Codex for OSS" program to offer maintainers free access and support.