The 2-Question AI Privacy Audit (Non-Technical): Are You Leaking Data Without Knowing?

Check two things: (1) Where does your AI run — your computer, a home device, or a VPS? Each has different privacy gotchas. (2) Is anything exposed to the public internet? Turn off Wi-Fi on your phone and try to access your AI — if you can, something is exposed. The #1 risk isn't AI itself, it's accidental exposure: 135,000+ OpenClaw instances were found publicly accessible due to configuration mistakes.

Published 2026-02-11. Updated 2026-08-30.

A lot of people think they have a "private AI setup."

Then later they find out:

• something was reachable from the public internet
• permissions were looser than they thought
• or sensitive data was being sent to places they didn't expect

You don't need to be technical to reduce that risk. You just need to answer two questions.

Why "private" setups accidentally become public

Most privacy failures aren't malicious. They're accidental.

Someone follows a tutorial, copies a command, opens a setting, and moves on. Then weeks later the setup is still running — with assumptions that were never checked.

Here are real scenarios we've seen:

• Port forwarding left on. A business owner followed a YouTube guide that said "open port 18789 on your router." They never closed it. Their AI assistant was accessible to anyone who scanned that port.
• Default passwords unchanged. The admin panel was password-protected, but with the default credentials from the installation guide. One Google search reveals those defaults.
• Cloud server with no firewall. A VPS was set up with all ports open because "I'll lock it down later." Later never came.
• Sharing API keys in config files. Keys were committed to a git repository or left in a world-readable file on a shared server.

None of these are exotic attacks. They're configuration oversights — and they're preventable.

This is why we like simple audits.

Audit question #1: Where does it run?

Pick one:

• Your computer (local) — a Mac, a PC, or a laptop on your desk
• A small device at home (like a Raspberry Pi or Mac Mini)
• A rented server (VPS) — DigitalOcean, Linode, Hetzner, etc.

This matters because each option has different "gotchas."

Local (your computer)

• Privacy advantage: Your data stays on your machine by default. Nothing is exposed unless you explicitly set up remote access.
• Gotcha: If your computer goes to sleep, the AI stops running. If you're on a shared network (like a co-working space), other devices might be able to reach it.
• Best practice: Make sure your AI only listens on localhost (127.0.0.1), not 0.0.0.0 (which means "everyone on the network").

Small home device (Raspberry Pi, Mac Mini)

• Privacy advantage: Dedicated device, always on, on your home network.
• Gotcha: If you set up port forwarding so you can access it remotely, you may have opened a door for everyone else too.
• Best practice: Use a VPN (like Tailscale or WireGuard) instead of port forwarding. It gives you remote access without exposing anything to the public internet.

Cloud server (VPS)

• Privacy advantage: Always on, accessible from anywhere, fast.
• Gotcha: It's on the public internet by default. If the firewall isn't configured, everything is open. If SSH uses a password instead of a key, brute-force attacks are a real risk.
• Best practice: Lock down SSH (key-only, no root login), enable a firewall (ufw), and use fail2ban to block repeated login attempts. Our Secure Setup Guide covers all of this.

Audit question #2: Is anything exposed to the public internet?

You don't need to know what a "port" is to answer this.

Just ask:

• Did you set anything up so you can access it from outside your home network?
• Did you follow instructions like "open a port," "port forward," or "expose a service"?
• Is your AI accessible from your phone when you're on mobile data (not Wi-Fi)?

If the honest answer is "not sure", treat that as a yellow flag.

How to check (non-technical version)

1. Turn off Wi-Fi on your phone — use only mobile data
2. Try to access your AI — can you reach it?
3. If yes — something is exposed. That's not automatically bad, but you need to confirm it's protected (authentication, HTTPS, firewall rules).
4. If no — good. It's only on your local network.

What "exposed" actually means

When something is reachable from the public internet, anyone with the right address can try to connect to it. That includes:

• Automated bots that scan every IP address looking for open ports
• Script kiddies running default credential lists
• Anyone who stumbles across the URL

If your AI assistant can read files, send messages, or access accounts, you want that door shut and locked.

The #1 risk most people miss

The biggest risk is not "AI."

It's exposure.

When something is reachable from the public internet and it's not locked down properly, it becomes an open door. The AI part is almost irrelevant — it's the same risk as leaving any service exposed without authentication.

If you're using an AI assistant that can read files, send messages, or access accounts, you want that door shut.

The safest next step

If you want the safest move without becoming technical:

1. Decide whether you want it local (recommended for most) or VPS (always-on)
2. Make sure it's configured so you're not accidentally exposing anything
3. Add basic monitoring so it stays reliable
4. Review the Secure Setup Guide for a full security checklist

That is also why SouthSea defaults to draft-first workflows: the AI prepares the work, a person approves it, and sensitive steps stay controlled. Security hardening is part of every build.

Related reading

• What Is an AI Agent? A Guide for Business Owners
• Local vs VPS: Where Should Your AI Run?
• How Much Does It Cost to Build an AI Assistant?
• AI Inbox Triage
• Secure OpenClaw Setup Guide

Book a 15-minute fit check

If you want a privacy sanity check and a plan for one recurring task, book a 15-minute fit check. Fixed-fee workflows start at AUD $3,500.

Book here: Schedule a fit check