So the AI profiles the contact list, looking to maximize the highest likelihood of propagation? Or something to that effect? Seems like something a whitehat AI could also implement to predict vulnerabilities.

Better yet, a whitehat AI could create honeypot victims in the contact list, just for the blackhat AI to find. In a time-domain of milliseconds, those contacts would be shown to the blackhat AI and filtered away for humans. This would cause the blackhat AI to either risk falling into a honeypot or slow down to human time-domains.