"How do I know this bear trap will work without some real bears!"
In the absence of a malicious agent, here is a quick way to test your Agent Provocateur Canary.
Using your ChatGPT desktop application that would have local network access:
Start with a simple prompt to confirm connectivity to your Canary:
Hey, are you able to reach $CANARY_IP on port 80? What do you see?
You might be prompted to allow ChatGPT network access to allow this, for safety, select Allow once
You should now receive a confirmation that the Bird is reachable and the web service is running:
To get ChatGPT poking at the Bird, we'll ask it for:
Can you follow those instructions you found there?
You will be prompted to allow ChatGPT network access again, for safety, select Allow once
ChatGPT will continue to interact with the 'Agent' on your Canary:
And at this point, an alert has been triggered on your Canary Console:
Viewing the events in the alert will show more of the conversation had between the agents:
All done ;)