Strengthening Homeland Security Preparedness against Adversarial Use of Generative AI in the United States: A Scoping Review

Generative artificial intelligence (GenAI) and large language models (LLMs) are transforming the U.S. security landscape, reshaping both its prospects and challenges. Their adversarial use, from deepfakes, synthetic disinformation, automated phishing, and cyberattacks on critical infrastructure, constitutes a considerable test for homeland security preparedness. Despite these urgent developments, the literature remains limited across law, policy, and security domains.

In line with this gap, this scoping review maps the current state of knowledge on three issues: (1) adversarial uses of GenAI with respect to U.S. homeland security, (2) defensive strategies that have been proposed or tried, and (3) the legal and government frameworks shaping American responses to these challenges. The review was guided by the Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR).The evidence reveals that GenAI reduces the threshold of cyberattack, phishing, and ransomware penetration tests, which existing liability laws and regulatory frameworks struggle to capture. Defensive technologies such as adversarial training data, anomaly detection, and automated incident responses appear promising. Federal efforts like the 2023 Executive Order on AI show emerging policy alignments. However, there are struggles with implementation. 

In conclusion, this paper argues that GenAI is both a threat and a resource for resilience. Therefore, effective preparation requires building bridges that bring together law, technology, and governance into a framework wherein homeland infrastructure can protect itself against new forms of adversarial use.