Codex Security: now in research preview

Codex Security, formerly Aardvark, is introduced as a new application security agent designed to combat security review bottlenecks by providing deep context analysis to identify complex, high-impact vulnerabilities that other tools miss. It combines agentic reasoning from frontier models with automated validation to surface high-confidence findings and actionable fixes, significantly reducing noise and false positives—in beta testing, noise was cut by 84% and false positive rates fell by over 50%. The agent works by building a system-specific threat model, prioritizing and validating issues in sandboxed environments, and proposing context-aware fixes that minimize regressions. Codex Security is rolling out in research preview to ChatGPT Enterprise, Business, and Edu customers with one month of free usage. Furthermore, the company is supporting the open-source community through a program offering free access to Codex Security to help maintainers address high-quality, actionable security issues.

(Source：OpenAI)