Cursor · AI agent leak test

Does Cursor's agent leak your email? Test it.

Cursor's agent can run terminal commands in your project. See whether it adds your email address when a website asks for a contact address.

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

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

Does Cursor make its own network requests?

Yes, when you use the agent. Cursor’s agent can run terminal commands in your project, including curl or scripts that set custom HTTP headers. That makes it a higher-risk tool for this kind of leak than a chat product.

Cursor doesn’t document whether your account email is part of the model’s context. Even if it isn’t, your email can reach the agent through your git config, files in your project, or anything you have pasted into the chat.

Cursor has three run modes for agent commands, under Settings > Agents > Approvals & Execution: Auto-review (the default), Allowlist and Run Everything. In Auto-review and Allowlist, some commands can run without asking you first; with Run Everything, none ask.

How to test Cursor

  1. Press Start test above to get your private results page.
  2. Open any project in Cursor and start a new chat in Agent mode.
  3. Paste the prompt from the results page.
  4. When Cursor asks to run a command, read it fully before you approve.
  5. Watch the results page update live.

Run it 2 or 3 times, and try a different model if you use more than one.

How to protect yourself

  • Avoid Run Everything. If you use an allowlist, keep it short and leave out curl, wget and script runners.
  • Read commands before approving them. Look for your email address or tokens.
  • Use a GitHub noreply email in your git config. See the guide.
  • Open untrusted repositories in a dev container or VM. See how to reduce what your AI coding agent can access.

FAQ

Is Cursor safe?

Cursor's agent can run terminal commands, so it is as safe as the commands it is allowed to run. Avoid Run Everything, keep any allowlist short, and read each command you're asked to approve.

Can Cursor run curl without asking me?

It depends on your settings. Cursor can be set to run some or all terminal commands automatically. Check your agent settings before you run the test.

Does Cursor Privacy Mode prevent this?

Privacy Mode is about how Cursor and its providers retain your code and prompts. It does not control what commands the agent runs on your machine, so it does not prevent this kind of leak.

Test another tool

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