The document, titled "Risks and Mitigation Strategies for Adversarial Artificial Intelligence Threats: A DHS S&T Study," prepared in June 2023, addresses the emerging risks associated with adversarial artificial intelligence (AAI) and suggests mitigation strategies. It explores how the Department of Homeland Security (DHS) can adapt to these challenges, especially in light of AI's transformative impact on various sectors. The report discusses different types of adversarial AI, such as evasion attacks, data poisoning, model extraction, and inference attacks, along with generative deceptive AI like deepfakes. Each of these threats is analyzed with respect to potential risks they pose and mitigation strategies. The document emphasizes the importance of understanding these risks early in the system lifecycle and integrating AI security assessments and standards into systems engineering processes. The study also highlights the need for international collaboration in developing standards and test methods to counter AAI threats effectively.

A copy of the document can be found here.