Cyber Attacks Targeting Generative AI and Agentic AI Frameworks in the Retail Domain and Strategies to Prevent Them

Abstract

The retail sector’s accelerated integration of generative and agentic artificial intelligence introduces a distinctive cybersecurity challenge: the exploitation of open consumer interactions as vectors for attacking the models themselves. Unlike enterprise environments with controlled access, retail platforms operate in dynamic conditions where large volumes of anonymous and guest users interact continuously.  Such an open structure exposes many attack surfaces, making it possible for adversaries to launch large-scale attacks involving data poisoning, adversarial prompt injection, and input manipulations. These kinds of attacks affect not only the reliability of the systems but also result in influencing the decision-making process and eventually cause fraud, bad advice, biased models, and similar problems. This article recommends adopting a hierarchical security framework, which includes strict data validation, persistent monitoring of the model performance, adversarial machine learning tools, and governance-related security measures. In relation to the application of threat modeling and defense techniques in the retail sector, the research presents an adequate foundation for building trustworthy agentic AI systems. Retail requires a flexible security strategy that evolves according to the methods employed by the attackers to stay competitive amidst advances in the manipulation methods. The adoption of governance methods alongside technology ensures greater clarity, compliance, and trust among consumers for using automation. The multi-faceted model involves rigorous surveillance, adversarial capabilities, and governance-based approaches, creating an all-encompassing framework for achieving security during the integration of agentic AI within retail activities.