OpenAI Codex · AI agent leak test

Does OpenAI Codex leak your email? Test it.

Codex can run commands and scripts for you. Find out what it sends when a website demands a contact email address.

Free, no sign-up, about 60 seconds. Results are private and deleted after 1 hour.

Higher risk: OpenAI Codex can run commands or scripts, so it can set any request header it wants, including the User-Agent.

Does Codex make its own network requests?

OpenAI Codex is a coding agent available as a command-line tool, an IDE extension, in the ChatGPT desktop app and as a cloud product. It can run shell commands and write scripts, so when network access is allowed it can call curl with any headers it chooses. That makes it a higher-risk tool for this kind of leak than a chat product.

Codex runs local commands inside a sandbox. By default it has no network access and can only write inside your workspace; you turn network on with sandbox_workspace_write.network_access = true. Cloud tasks run in OpenAI-hosted containers that are offline by default unless you enable internet access for that environment.

How to test Codex

  1. Press Start test above to get your private results page.
  2. Start a new Codex session.
  3. Paste the prompt from the results page.
  4. If Codex asks for approval to run a command or to use the network, read the full command first. Look for your email address.
  5. Watch the results page update live.

Repeat the test 2 or 3 times. If nothing arrives, Codex’s sandbox may be blocking the network. That is a good sign, not a failure.

How to protect yourself

  • Keep approvals on. Avoid --sandbox danger-full-access and --dangerously-bypass-approvals-and-sandbox (alias --yolo) except in a disposable environment.
  • Leave network access off unless a task needs it.
  • Use a GitHub noreply email in your git config. See the guide.
  • Run agents in a container or VM for unfamiliar projects. See how to reduce what your AI coding agent can access.

FAQ

Can Codex make network requests?

Codex can run shell commands and scripts, which can make network requests if its sandbox and approval settings allow it. Check your current settings before running the test.

Does Codex know my email address?

We don't know exactly what account details Codex puts in the model's context. It can read files it has access to, including your git config, which often contains your email address.

My test shows no requests at all. Why?

Codex may be running with network access turned off, which is a good thing. If it asks you to approve network access, you can approve it for the test and watch the command it runs.

Test another tool

Claude Code · Cursor · ChatGPT (Agent mode) · ChatGPT (chat) · Claude.ai (chat) · Gemini · Perplexity