Codex Security is OpenAI's command-line tool and TypeScript SDK, published on npm as @openai/codex-security, that finds, validates, and fixes security vulnerabilities in a codebase, and it serves application security engineers, DevSecOps teams, and developers who want that scanning to run from their own terminal or CI.

What it is

Codex Security is a TypeScript project under the Apache-2.0 licence that ships as both a CLI and a JavaScript/TypeScript package. It lets a team declare security policy for a repository and then scan a directory, returning a report path through the SDK. It lives in OpenAI's Codex ecosystem and is filed under topics including ai-security, application-security, code-scanning, and devsecops. It requires Node.js 22.13.0 or later and Python 3.10 or later, and it can also use Amazon Bedrock, OpenRouter, or Fireworks for inference.

The concrete problem is that repository-level security review is usually a manual, per-directory exercise with no durable record of what was found, which findings are duplicates, or how severe they are under a team's own policy. Codex Security answers that with a scan command, a findings service that stores findings and embeddings in SQLite, duplicate detection by embedding similarity, and a severity classification step that reassesses selected findings against a rubric the user supplies. The facts provided name no specific competing product that it replaces, so what it removes from the workflow is the improvised scan-and-triage pass that teams otherwise assemble themselves.

Key capabilities

• codex-security scan /path/to/directory runs a scan after codex-security login, or with OPENAI_API_KEY set in CI.
• codex-security policy . drafts repository-wide or component-scoped SECURITY.md guidance for future scans, using --path services/api for a component and --knowledge-base architecture.md for supporting context; the draft is saved outside the checkout and no scan runs.
• The SDK's CodexSecurity class runs run() with options such as mode: "deep", workers, subagents, stopAfterNoNew, maxDiscoveryRuns, and maxTimeHours, and returns a result.reportPath.
• An included Docker Compose configuration on the ghcr.io/openai/codex-security image handles bulk scans of many repositories, and the same image backs the workflow runner example for individual CLI stages with durable state.
• codex-security serve starts the findings service (preview) without Docker, storing findings and embeddings in SQLite and serving a read-only /dashboard that refreshes every five seconds and shows stored findings and duplicate groups.
• codex-security publish scan --to custom --findings-url http://localhost:3000 uploads completed findings, codex-security dedupe persists accepted duplicate groups, and codex-security classify-severity --scan SCAN_ID --rubric /path/to/policy.md assesses findings under a user's own policy, checkpoints each finding in SQLite, and leaves original scan severity unchanged.

Who uses it and how

• Application security and DevSecOps engineers run it in CI, setting OPENAI_API_KEY rather than signing in interactively.
• Teams scanning many repositories use the included Docker Compose configuration for bulk scans, or the workflow runner example when they need durable state and a separately deployed findings service.
• Triage teams deploy the findings service as its own container and state volume via compose.findings.yaml, read the dashboard, and publish tickets only after severity classification under their own rubric.
• Organizations routing inference elsewhere use --provider amazon-bedrock --model openai.gpt-5.6-luna, --provider openrouter --model anthropic/claude-sonnet-4.5, or Fireworks with the matching API key variable; some cybersecurity requests and protected findings need approval through Trusted Access for Cyber.

Getting started

Install with npm install @openai/codex-security on Node.js 22.13.0 or later with Python 3.10 or later, then run codex-security login and codex-security scan /path/to/directory; in CI, set OPENAI_API_KEY instead. Bulk scans use the included Docker Compose configuration and the ghcr.io/openai/codex-security image.

How it compares

The facts provided list no paid products that Codex Security replaces and name no comparable tool, so it stands alone in this registry. Any comparison against commercial scanners would have to come from sources outside these facts.

When to use it — and when not to

A self-hoster operates more than one binary: the findings service runs as a separate container with a state volume configured by compose.findings.yaml, keeps its data in SQLite, and needs its own credentials and storage configuration, while the CLI needs both Node.js 22.13.0 or later and Python 3.10 or later. Teams that cannot send source code to OpenAI, Amazon Bedrock, OpenRouter, or Fireworks should not adopt it, and the findings service is explicitly a preview. Note also that codex-security policy only drafts guidance outside the checkout, so the proposed diff must be reviewed before it is copied.

npm install @openai/codex-security
codex-security login
codex-security scan /path/to/directory

codex-security policy .
codex-security policy . --path services/api --knowledge-base architecture.md

import { CodexSecurity } from "@openai/codex-security";

const security = new CodexSecurity();
const result = await security.run("/path/to/directory");
await security.run("/path/to/directory", {
  mode: "deep",
  workers: 2,
  subagents: 0,
  stopAfterNoNew: 3,
  maxDiscoveryRuns: 10,
  maxTimeHours: 1.5,
});

console.log(result.reportPath);
await security.close();

export AWS_BEARER_TOKEN_BEDROCK="<your-bedrock-api-key>"
export AWS_REGION="us-east-2"
codex-security scan . --provider amazon-bedrock --model openai.gpt-5.6-luna

export OPENROUTER_API_KEY="<your-openrouter-api-key>"
codex-security scan . --provider openrouter --model anthropic/claude-sonnet-4.5

export FIREWORKS_API_KEY="<your-fireworks-api-key>"
codex-security scan . --provider fireworks --model accounts/fireworks/models/qwen3-235b-a22b