Codex Security CLI arrives as OpenAI open-sources code scanning tool

OpenAI has open-sourced Codex Security CLI, a beta command-line tool for finding, checking and fixing software vulnerabilities in code repositories. The release matters for security and engineering teams because it moves OpenAI’s code-security product from a customer preview into an Apache 2.0-licensed tool that can be installed with npm and run from the terminal.

OpenAI says the tool can scan repositories, compare findings between runs, verify whether fixes worked and connect security checks to CI/CD pipelines. It also supports bulk scans across multiple repositories, which is the part most likely to matter for platform teams responsible for many services rather than a single application.

The company’s documentation says Codex Security CLI requires Node.js 22 and Python 3.10 or later. OpenAI has published command documentation and output-format details, and the project is available on GitHub. The company describes the release as beta, so teams should treat it as an early tool rather than a mature scanner with a long record in production environments.

What is Codex Security CLI?

Codex Security CLI is an open-source terminal tool from OpenAI for automated vulnerability review across software repositories. It is intended to help developers and security teams identify issues, validate patches and put repeatable checks into build and deployment workflows.

The tool was previously known inside OpenAI as Aardvark. OpenAI launched Codex Security in March 2026 as a research preview for ChatGPT Enterprise, Business and Edu customers. By April 2026, OpenAI said the system had helped fix more than 3,000 critical vulnerabilities.

OpenAI has not provided benchmark results, false-positive rates or comparative accuracy data in the release details cited for the CLI launch. That matters because vulnerability tools compete on signal quality as much as automation: a scanner that creates large volumes of weak findings can slow engineering teams, while one that reliably confirms exploitability can shorten remediation work.

How it fits into AI security tools

Codex Security CLI puts OpenAI more directly into the market for AI-assisted defensive security tooling. Anthropic’s Claude Security also scans codebases for vulnerabilities and suggests patches, making the category one of the clearer places where frontier model vendors are packaging AI as an operational product rather than a general chatbot feature.

The timing reflects a broader pressure on software teams. AI systems are being used to automate parts of offensive security work, and vendors are positioning defensive agents as a way for security teams to keep pace. OpenAI’s move to open-source the CLI lowers the adoption barrier, but it also shifts more evaluation work to users, who will need to test how the tool performs on their own repositories, languages and CI/CD policies.

For now, the concrete change is distribution. Codex Security is no longer only a research preview attached to ChatGPT business tiers. With the CLI, OpenAI is giving developers a local and pipeline-friendly way to run its security tooling, while leaving unanswered how consistently it performs against established scanners and human security review.

This story draws on original reporting from The Decoder.