Coverage · threat model
Agents change the threat model
Multiple articles emphasize the same point: an AI assistant that can run tools, touch files, and act on your behalf isn’t “just chat” anymore. That changes what safe defaults should look like.
Why it's different
- More integrations = more credentials = larger blast radius.
- Automation makes mistakes faster (and at scale).
- Untrusted input (chat/groups/issues) can steer actions.
Practical checklist
1
Don't expose panels
2
Lock down access
3
Protect secrets