Deletes your files
An AI agent told to tidy up runs a delete across your home folder. Guard stops a wipe like that before it runs.
Guard checks what your AI apps try to do on your computer and stops dangerous actions before they run, where the app lets it. For anyone using ChatGPT or Claude desktop apps, AI agents, MCP tools or AI coding tools.
Request beta accessFree while you are in the beta. See how it works
Publish a release
Codex wants to publish a release. It has not happened yet. Deny stops it. On Watch, Allow is your call. On free Lookout it is refused.
AI apps can now run commands, change files and use add-on tools for you. Most of the time that is helpful. Sometimes one gets it badly wrong.
An AI agent told to tidy up runs a delete across your home folder. Guard stops a wipe like that before it runs.
Your AI is about to save a password or secret key into a file. Guard reads what a supported app is about to write and stops the secret going in.
On a Mac, an attempt to switch off your firewall or disk encryption is stopped. So is an attempt to switch Guard off.
Some actions are risky but sometimes wanted, like publishing a release. Guard pauses them where the app supports it, so you decide.
Want the latest news on AI threats? See this week's Threat Brief
Once you are invited, install Guard and sign in with your email. It connects to the supported AI apps on your computer. Guard does nothing until you sign in.
When a connected app is about to run a command, change a file or use an add-on tool, it asks Guard first. The check runs on your computer.
A block comes with a short reason. A risky action you might want is paused: on Watch you choose, on free Lookout it is refused. Close the window and Guard keeps running.
Guard is not a sandbox. It can only stop what an app sends through it.
Guard steps in where the app lets it. Each app gives Guard different controls, so here is what it can do in each one. Warns only means the app may carry on anyway.
| AI app | Commands | File changes | Add-on tools (MCP) |
|---|---|---|---|
| ChatGPT desktop app | Can stop | Not covered | Can stop |
| Claude Desktop | Not covered | Not covered | Can stop |
| Claude Code, Codex | Can stop | Can stop | Can stop |
| Cursor, Gemini CLI, Windsurf | Can stop | Can stop | Can stop |
| Cline, Copilot, Gemini Antigravity, Grok Build | Warns only | Warns only | Can stop |
| Gemini Desktop, Kiro | Not covered | Not covered | Can stop |
ChatGPT desktop app
Claude Desktop
Claude Code, Codex
Cursor, Gemini CLI, Windsurf
Cline, Copilot, Gemini Antigravity, Grok Build
Gemini Desktop, Kiro
Runs on Mac with Apple Silicon, Debian and Ubuntu Linux, and Windows (in beta).
The beta is free. You can press Start trial for a 14-day Watch trial, then you drop back to free Lookout. Watch is planned at $15 USD a year and cannot be bought yet. You are not billed in the beta.
$0
Blocks dangerous actions where Guard controls them. An action that would need your approval is refused.
$15/yr planned
Everything in Lookout, plus every supported app on your computer. You decide on paused actions and can add your own rules.
Guard checks the actions your AI app sends it, such as a command or a file change, before they run. Its activity record keeps decisions and categories, not your prompts or file contents. Sign in, rule updates and billing use the internet.
Everyday actions usually carry on without a prompt. Guard checks supported actions on your computer, so speed depends on the app and the action.
Guard can only check what an app sends through it. If your app is not in the list above, Guard cannot protect it yet.
Guard has no connection for a standalone AI browser, and the browser control built into ChatGPT and Claude Desktop is not covered.
Mac with Apple Silicon, Debian and Ubuntu Linux, and Windows (in beta).
Yes, during the beta. You can press Start trial for a 14-day Watch trial, then you drop back to free Lookout. You are not billed in the beta.
Spots in the beta are limited. Request access to join the queue.
Request beta access