0:00 / 0:00

Make Your AI Ask Before It Acts

From the course: Agentic AI For Founders & Senior Operators

Open each connector in Claude to see its tools and what each one is allowed to do. Set anything you don't want done without you, such as sending an iMessage, to need your approval. As a rule, anything that sends should need approval, or the AI will send things on your behalf without you knowing.

Transcript

The other way you alter settings at an individual permission level is you would go into your connectors here, and you can go into each connector. If you click into this, it will show you all the tools and what their current allow list is. So, you know, things like Canva, I don't use much, so it's on Always allow, but you may have different permissions that work for you. So, for instance, you've got something that, you know, is an MCP, and, you know, you don't want it to do anything without your explicit approval. So, in this case, for instance, I've said, "Hey, if you want to send an iMessage, you need explicit approval from me." Typically, anything that is send-related, you want explicit approval because otherwise it will take actions and you won't know what's happening. It will also send things on your behalf, right? Again, this should be editable by you if you come into this menu right here.