How is the performance of our generated rules in the wild?
To demonstrate the effectiveness of the rules generated by our workflow in the wild, we deploy a honeypot to safely attract and capture security event logs from real-world attacks.
We mainly collect the web application security logs to demonstrate the generalization effectiveness of diverse types of security logs/events.
Specifically, we follow the DShield, which uses a Cowrie honeypot on an Amazon Web Service (AWS) server with the 3-month free plan, to attract the general web attacks in the wild.
In summary, our honeypot collects web application logs with a total of 64,185 events for three months from 22nd Jul-22nd Oct 2024.
We report the percentage of precision and recall in the wild across three rule sets, where we can find that the rules generated by \name significantly outperform existing Sigma or Splunk rules in detecting malicious events in the wild.