Risks and Mitigation Strategies for Adversarial Artificial Intelligence Threats: A DHS S&T Study

As the artificial intelligence (AI) landscape evolves, the pursuit of smarter algorithms has given rise to an AI-based sub-discipline that blurs the lines between innovation and deception: adversarial AI (AAI), where AI systems not only make predictions and take actions, but can also engage in a strategic dance of deception and counter-deception. This deception can target both humans as well as AI-based systems, thus breaking the reliability of the AI systems themselves and shaking the faith we have in the digital content we consume. This report introduces adversarial AI concepts and explores future AAI threats, risks, and mitigation strategies—to help the Department develop a risk-informed approach to mitigating AAI threats and vulnerabilities.