MIT Logo

Poison, Persistence, and Cascade Effects: AI and Cyber Conflict

Abstract:

“The rise of AI-augmented cyber defenses incorporated into national defense postures will likely be vulnerable to “poisoning” attacks
that predict, manipulate, and subvert the functionality of defensive algorithms. These AI-enabled cyber campaigns contain great potential for operational obfuscation and strategic misdirection. At the operational level, piggybacking onto routine activities to evade security protocols adds
uncertainty, complicating cyber defense particularly where adversarial learning tools are employed offensively. Strategically, AI-enabled cyber
operations may be able to pursue conflict outcomes beyond those expected of adversaries. Perhaps more worrisome is that the centrality of the Internet to new AI systems incorporated across all areas of national security—not just to cyber conflict processes—indicates that sophisticated adversaries may be motivated to launch offensive online actions to achieve effects in other domains with some increasing regularity.”

Author:
Christopher Whyte
Year:
2020
Domain: ,
Dimension:
Region: ,
Data Type: ,
MIT Political Science
MIT Political Science
ECIR
GSS