● What is open: the Codex Security CLI and SDK, under Apache 2.0.
● What is not: the analysis service behind it stays under OpenAI’s control. The client is public, the engine is not.
● Who can use it: limited beta — approved customers and partners only, for now.
● The useful bit: it reviews staged and unstaged changes before a commit, not just on the branch.
OpenAI has open-sourced the command-line interface and SDK for Codex Security under Apache 2.0. The tool scans repositories, reviews changes before they are committed, compares findings between scans so you can see what a change introduced rather than re-reading the whole list, and exports results in a form other code-scanning tools can consume.
In continuous integration it can fail a check when a completed scan finds an issue at or above a severity level you choose — which is the part that makes it a merge gate rather than a report nobody reads.
The split is worth being precise about. Apache 2.0 on the client means you can read it, fork it, and audit what leaves your machine. It does not mean you can run the analysis yourself: that service remains OpenAI’s, and the CLI and SDK are in limited beta restricted to approved customers and partners. Calling the product open source without that caveat would be wrong.
Related: which coding agent vendors actually shipped security fixes.
Sources