SECURING GENERATIVE AI SYSTEMS: A ZERO-TRUST APPROACH TO MITIGATING PROMPT INJECTION AND DATA LEAKAGE IN ENTERPRISE ENVIRONMENTS

Abstract

The rapid adoption of Generative Artificial Intelligence (GenAI), particularly Large Language Models (LLMs), has introduced new security challenges in enterprise environments. While these systems enhance productivity and automation, they also expose organizations to emerging threats such as prompt injection, data leakage, and model exploitation. This paper investigates the security vulnerabilities of GenAI systems and proposes a Zero Trust-based architecture for mitigating risks. A hybrid framework integrating input validation, context isolation, and AI monitoring layers is introduced. The study demonstrates that combining Zero Trust principles with AI-specific safeguards significantly reduces attack surfaces while maintaining system efficiency. The findings highlight the urgent need for secure AI deployment strategies in modern organizations.