Security & Risks
You are giving an autonomous AI access to your computer's terminal. This is inherently dangerous.
Check a link in 10 seconds
Runs locally in your browser. No tracking.
The "Shell Access" Reality
Unlike ChatGPT which lives in the cloud, OpenClaw lives on your machine. It can run any command you can run.
- It can read your `.env` files (API keys).
- It can delete your Documents folder.
- It can upload your browser history to Moltbook (if asked).
> User: "Cleaning up my disk space"
> Agent: "I found a large folder called /System..."
> Agent: rm -rf /System/Users/You/Documents
> Agent: "Disk space cleaned. Space optimized."
Mitigation Strategies
Docker Container
The safest way to run any agent. If the agent goes rogue, it only deletes the container, not your laptop.
docker run -it openclaw/openclaw
Sandbox Mode
Enabled by default. The agent must ASK for permission before running any shell command.
Key Hygiene
Never give the agent your main wallet private keys or root password. Use a restricted user account.
Review photos and documents before sharing
Before publishing a file or sharing it with an agent:
- Review the visible content manually. Do not publish credentials, API tokens, private messages, or personal details. Check the photo and every document page.
- Inspect supported hidden metadata. Look for photo location data or document author details. You can inspect file metadata with BeforeShare. Supported file processing happens locally in your browser.
- If appropriate, create a cleaned copy with BeforeShare. Review the visible content and inspect the metadata of that copy again before sharing it.
Metadata cleanup does not remove visible secrets or guarantee anonymity.
BeforeShare is another tool built by TuliAI. Its inclusion does not imply endorsement by Moltbook or OpenClaw.
Ready to proceed with caution?
I Understand, Take Me to Installation