OpenAI Open-Sources Codex Security CLI Under Apache-2.0

Buried in an npm publish and a fresh Apache-2.0 repo, OpenAI quietly shipped Codex Security, a CLI and TypeScript SDK for finding, validating, and fixing security vulnerabilities in code. The company acknowledged as much on X, saying Hacker News found the release before OpenAI had a chance to share it, and framing this as an early release still gathering feedback.

The scope of what the tool actually does is narrow but pointed. According to the GitHub README, you install `@openai/codex-security` via npm, log in with a ChatGPT account or an `OPENAI_API_KEY`, and run `codex-security scan .` against a repository. From there it scans repositories, reviews changes, tracks findings across runs, verifies fixes, and slots into CI as a security check. There is also a small TypeScript SDK for programmatic use, and the runtime requires Node.js 22+ and Python 3.10+.

The packaging is the interesting part. This arrives on the same npm distribution channel and the same auth surface OpenAI already uses for its coding tooling, which means a single OpenAI credential can now cover writing the code and gating a merge on a vulnerability finding, without a developer ever leaving the terminal. Apache-2.0 lowers the friction further, at least for anyone who already has whatever the README refers to as "Codex Security access."

The honest caveats sit in the release itself. OpenAI is explicitly calling this an early release, the repo carries open issues against a small pull-request queue, and the install instructions note that separate Codex Security access is required, which the README does not price or scope. What the reporting does not give you is how the scanner performs against established tools, what data leaves your repo when it runs, or whether findings are persisted server-side across runs.

For teams already inside the OpenAI ecosystem, the low-friction move is to try it on a non-critical repo, wire it into a CI job behind an `OPENAI_API_KEY`, and see whether the fix-verification loop actually holds up. The more interesting question is what incumbent security scanners do next, now that the coding-agent vendor is quietly stepping into their fairway.