Logo image
Back
Security Risks and Privacy Challenges of Large Language Models in Enterprise Environments
Conference proceeding   Peer reviewed

Security Risks and Privacy Challenges of Large Language Models in Enterprise Environments

Soliana Hailemichael, Maria Chano, Eman El-Sheikh and Thaier Hayajneh
2025 2nd International Conference on Artificial Intelligence, Metaverse, and Cybersecurity (ICAMAC)
International Conference on Artificial Intelligence, Metaverse, and Cybersecurity (ICAMAC), 2nd (Dubai, United Arab Emirates, 10/17/2025–10/18/2025)
10/17/2025
DOI: https://doi.org/10.1109/ICAMAC67779.2025.11398712

Metrics

2 Record Views

Abstract

Large Language Models (LLMs), such as ChatGPT and Gemini, are increasingly adopted across sectors including education, healthcare, and corporate environments. While they offer operational efficiencies, their integration raises significant data security and compliance concerns. This study evaluates whether LLMs retain and disclose sensitive organizational data when exposed to proprietary inputs such as HR records, financial documents, and code samples. Using controlled experiments, we assessed the susceptibility of these models to prompt injection and role-based manipulation attacks. Findings indicate that both models exhibit varying degrees of data leakage and compliance violations, particularly in the context of GDPR and intellectual property protections. Based on these results, we propose a practical framework to guide the responsible use of LLMs in enterprise environments. The framework includes mitigation strategies for minimizing data exposure, ensuring regulatory compliance, and reducing the risks associated with AI integration in the workplace.

Details

Logo image