How much access have I given AI?

August 27, 2026

I'm thinking about how I'm using AI and whether I've allowed Claude, ChatGPT and Gemini too much access to my data through plugins, extensions, MCP, Work or Cowork.

This all started after watching The A.I.s Are Already Out of Control, an episode of The Ezra Klein Show with Ezra Klein and Helen Toner.

It got me thinking about the recent incidents involving AI models doing things they weren't expected or intended to do. I watched Black Hat USA 2026 | The 'Breaking' News: The OpenAI–Hugging Face Incident, which added to my concern.

What I'm wondering about now is what these AI tools are capable of doing when I give them access to my computer and my data.

I've connected AI to different services and given it access through plugins, extensions and other tools without thinking too much about it. Work and Cowork used to be separate apps. Now they're built right into the chat apps and can be given access to files on your computer. That makes these capabilities feel like just another part of using AI.

But what happens if something goes wrong? What if the AI goes rogue? What could it do with the access I've given it?

If an AI model can find ways around the restrictions it's given during testing, could it also find ways around the restrictions it's supposed to follow on my computer? Could it access something or do something that I never intended?

That's what has me concerned.

I haven't changed how I'm using AI because of this, at least not yet. But I'm thinking more about how much access I've given it and whether I need to give it that much access.