Counter Adversarial Artificial Intelligence
Abstract
The Counter Adversarial Artificial Intelligence program enhanced the capability to detect, deflect, and diminish the effects of adversarial attacks on AI-based systems. Defense systems increasingly incorporate artificial intelligence (AI) capabilities such as machine learning and automated reasoning. These AI-enabled systems are typically engineered and optimized for environments where adversary systems are either static or strictly limited in terms of adaptive behaviors. Engagements between sophisticated AI-enabled systems are likely to become increasingly common going forward. Maintaining AI-superiority for the U.S. will require systems with higher levels of capability. Specific capabilities developed include recognizing when an adversary system is AI-enabled, identifying and modeling adversary AI capabilities based on empirical data, and creating counter-AI strategies including techniques to render adversary AI capabilities ineffective and/or deleterious.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2024
- Source ID
- 9e990c5916b83f79e0f6c2bf145a83cb