4.2 · What's Next
AI is powerful. It's also imperfect. Here's what you need to know about using Cowork safely, without the scare tactics.
Cowork is transparent about its work. Every action it takes shows up in your conversation. You can see exactly which files it read, what it created, and what changes it's proposing. There's no hidden activity.
When it wants to edit an existing file, it asks first. You see the proposed changes and approve or reject them. It doesn't silently overwrite your work.
Always review important output. Cowork is very capable, but it can make mistakes. If you're sending a client email, check it. If it's calculating numbers, verify the key figures. The same way you'd review work from a new team member before it goes out the door.
Be aware of what you share. When you point Cowork at a folder, it can see everything in that folder. If there are sensitive files you don't want the AI to process, put them in a separate folder. Cowork only sees what you point it at.
AI output is a draft, not a final product. For internal notes, summaries, and first drafts, Cowork's output is usually good enough to use directly. For anything external-facing or high-stakes, treat it as a strong starting point that you refine.
The game has a Level 3 demo about a newsletter with a hidden payload. It's a practical lesson about how AI can catch things humans miss, and also about why healthy scepticism matters. Worth playing through.
Cowork is a tool. Like any tool, it works best when you understand what it's good at and where it needs supervision. You wouldn't send a first draft from a new hire to a client without reading it. Same idea here.
The good news: Cowork shows you everything it's doing. You're never in the dark about what's happening. That transparency is built in, not bolted on.
Play the Level 3 security demo (Demo 9). It's a quick, practical lesson about staying sharp when working with AI.