OpenAI just dropped Codex Security, an application security agent that actually understands your project’s context. Instead of drowning you in false positives like most AI security tools, it finds real vulnerabilities by building a threat model of your system. It started as a private beta called Aardvark, where it caught critical issues like an SSRF and a cross-tenant auth flaw. The team slashed noise by 84% on one repo and cut over-reported severity by 90%+. Basically, it validates findings in sandboxed environments so you only see what matters. It’s rolling out to ChatGPT Pro, Enterprise, Business, and Edu users for free this month. Why it matters: As AI accelerates development, security review becomes the bottleneck—tools that ship high-confidence fixes could finally let teams move fast without breaking things.