Most people's first instinct after installing OpenClaw is: start playing with it.
Mine too. But later I realized — if you don't think through the security architecture from the start, things get harder and harder to clean up later. Your AI will have access to your data, your accounts, your behavioral patterns. Once these are exposed in an insecure environment, when something goes wrong, you won't even know how it went wrong.
So my principle is: it's not about keeping AI away from sensitive data — it's about building the guardrails before you let it touch anything.
This is the most basic step, and also the one most people are too lazy to do.
My approach: all AI-related operations use a dedicated, separate Google account. Not your daily-use account, not your work account, not the one you use to receive emails. A brand new account, exclusively for your AI system.
Why? Because AI operations require authorizing various services. If you use your main account, you're essentially handing over the keys to everything you have. A separate account means: even in the worst-case scenario, the damage is contained within that account's scope.
Practical tips:
My Tech Lead has a dedicated work environment — a Sandbox.
What does this mean? Before performing any risky operation, it runs it first in an isolated environment. For example, when I want to install a new skill pack, the process is: