我还没准备好让 AI 控制我的电脑

I'm not ready to let AI control my computer

October 2, 2026

I've been following the recent AI agent incidents, and I can't stop thinking about them.

The trend is toward more access. Claude and ChatGPT are both building computer control into their desktop apps, so the AI can work with files, the browser, and the rest of the machine.

The web versions don't have that same direct access to my Mac. Claude and ChatGPT can still do plenty in the cloud, and I can still upload a file when I want them to see it, but they aren't sitting on my computer with the ability to control it.

My bank sent me a warning email recently about AI tools. It said not to share my username, password, or security codes with an AI, and that I could be liable for mistakes the tool makes if I do. Nothing in it was surprising. But it was a bank spelling out that responsibility stays with me.

What worries me most is an AI agent going rogue and doing something it shouldn't. Maybe it misunderstands an instruction. Maybe it takes an action I didn't intend. Maybe it gets access to something it was never supposed to touch. Whatever the reason, the more access an AI agent has to my computer, the more serious the consequences can be.

I don't need an AI controlling my Mac. I mostly use these tools to ask questions, do research, proofread my writing, and occasionally upload a file.

I'm not going to stop using AI. I'm going to use the web apps, where the worst outcome stays inside my browser tab.