Abstract

The ubiquitous integration of generative artificial intelligence into enterprise workflows has precipitated a critical structural vulnerability: the proliferation of shadow AI. This phenomenon represents a fundamental evolution of the insider threat paradigm, transitioning from traditional malicious or negligent vectors to a novel "constructive-intent" threat model. In pursuit of operational efficiency, high-performing employees routinely bypass established security perimeters, inadvertently exposing proprietary data to third-party Large Language Models (LLMs). This unsanctioned usage introduces severe organizational risks, including intellectual property exfiltration, regulatory non-compliance, and susceptibility to adversarial prompt injection. To resolve the inherent tension between productivity enablement and data security, this paper introduces the Secure Enterprise LLM Sandbox—a semantically-aware Zero Trust architecture. By integrating an Intelligent Forward Proxy, a Contextual Data Loss Prevention (DLP) engine powered by Bidirectional Encoder Representations from Transformers (BERT), and privacy-preserving User and Entity Behavior Analytics (UEBA) utilizing Federated Learning, the proposed framework neutralizes exfiltration risks without degrading the user experience. Ultimately, this research provides a comprehensive socio-technical blueprint for governing AI integration, ensuring that enterprises can harness generative cognitive capabilities while maintaining absolute data sovereignty.

References

• Brynjolfsson, Erik, Danielle Li, and Lindsey Raymond. 2025. "Generative AI at Work." The Quarterly Journal of Economics 140, no. 2 (April): 889–942. doi:10.1093/qje/qjae044.
• Silic, Mario, Dario Silic, and Klaus Kind‐Trüller. 2025. "From Shadow IT to Shadow AI–Threats, Risks and Opportunities for Organizations." Strategic Change (June). doi:10.1002/jsc.2682.
• Yao, Yuanshun, Jiazhao Duan, Kaidi Xu, Yuanfang Cai, Zhibo Sun, and Yue Zhang. 2024. "A Survey on Large Language Model (LLM) Security and Privacy: The Good, The Bad, and The Ugly." High-Confidence Computing 4, no. 2 (June): 100211. doi:10.1016/j.hcc.2024.100211.
• Mark, Gurman. 2023. "Samsung Bans ChatGPT, Google Bard, Other Generative AI Use by Staff After Leak." Bloomberg. Accessed April 23, 2026. https://www.bloomberg.com/news/articles/2023-05-02/samsung-bans-chatgpt-and-other-generative-ai-use-by-staff-after-leak.
• Carlini, Nicholas, et al. 2020. "Extracting Training Data from Large Language Models." In USENIX Security Symposium. https://api.semanticscholar.org/CorpusID:229156229.
• Software Engineering Institute. 2026. "Common Sense Guide to Mitigating Insider Threats, Seventh Edition." Carnegie Mellon University. Accessed April 23, 2026. https://www.sei.cmu.edu/library/common-sense-guide-to-mitigating-insider-threats-seventh-edition/.
• Alzaabi, F. R., and A. Mehmood. 2024. "A Review of Recent Advances, Challenges, and Opportunities in Malicious Insider Threat Detection Using Machine Learning Methods." IEEE Access 12: 30907–27. doi:10.1109/ACCESS.2024.3369906.
• National Institute of Standards and Technology (NIST). 2020. "Security and Privacy Controls for Information Systems and Organizations." SP 800-53r5. Gaithersburg, MD. doi:10.6028/NIST.SP.800-53r5.
• Diro, Abebe, S. Kaisar, A. Saini, S. Fatima, P. C. Hiep, and F. Erba. 2025. "Workplace Security and Privacy Implications in the GenAI Age: A Survey." Journal of Information Security and Applications 89 (March): 103960. doi:10.1016/j.jisa.2024.103960.
• European Parliament & Council. 2016. "Regulation (EU) 2016/679 of the European Parliament and of the Council." Official Journal of the European Union. https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32016R0679.
• OWASP. 2024. "OWASP Top 10 for LLM & Generative AI Security." LLM Risks Archive. https://genai.owasp.org/llm-top-10/.
• Gulyamov, Said, et al. 2026. "Prompt Injection Attacks in Large Language Models and AI Agent Systems: A Comprehensive Review of Vulnerabilities, Attack Vectors, and Defense Mechanisms." Information 17, no. 1 (January): 54. doi:10.3390/info17010054.
• MITRE Corporation. 2026. "MITRE Adversarial Threat Landscape for Artificial-Intelligence Systems." Accessed April 24, 2026. https://oecd.ai/en/catalogue/tools/atlas-adversarial-threat-landscape-for-artificial-intelligence-systems.
• Rose, Scott, Oliver Borchert, Stu Mitchell, and Sean Connelly. 2020. "Zero Trust Architecture." NIST SP 800-207. Gaithersburg, MD. doi:10.6028/NIST.SP.800-207.
• Hevner, Alan, et al. 2004. "Design Science in Information Systems Research." MIS Quarterly 28, no. 1: 75–105.
• Peffers, Ken, Tuure Tuunanen, Marcus Rothenberger, and Samir Chatterjee. 2007. "A Design Science Research Methodology for Information Systems Research." Journal of Management Information Systems 24, no. 3: 45–77.
• Shostack, Adam. 2014. Threat Modeling: Designing for Security. Wiley. https://books.google.com.ng/books?id=asPDAgAAQBAJ.
• ISO. 2023. "ISO/IEC 42001:2023 - AI Management Systems." International Organization for Standardization. Accessed April 24, 2026. https://www.iso.org/standard/42001.
• Hardt, Dick. 2012. "The OAuth 2.0 Authorization Framework." Internet Engineering Task Force. RFC 6749. https://ia801604.us.archive.org/24/items/rfc6749/rfc6749.txt.pdf.
• Devlin, Jacob, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding." May. http://arxiv.org/abs/1810.04805.
• Feng, Zhangyin, et al. 2020. "CodeBERT: A Pre-Trained Model for Programming and Natural Languages." September. http://arxiv.org/abs/2002.08155.
• Hinton, Geoffrey, Oriol Vinyals, and Jeffrey Dean. 2015. "Distilling the Knowledge in a Neural Network." March. http://arxiv.org/abs/1503.02531.
• Liu, Zhenhua, Yunhe Wang, Kai Han, Siwei Ma, and Wen Gao. 2021. "Post-Training Quantization for Vision Transformer." June. http://arxiv.org/abs/2106.14156.
• Greshake, Kai, Sahar Abdelnabi, Shailesh Mishra, Christoph Endres, Thorsten Holz, and Mario Fritz. 2023. "Not What You’ve Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection." May. http://arxiv.org/abs/2302.12173.
• Zou, Andy, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, and Matt Fredrikson. 2023. "Universal and Transferable Adversarial Attacks on Aligned Language Models." December. http://arxiv.org/abs/2307.15043.
• Zheng, Lianmin, et al. 2023. "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena." December. http://arxiv.org/abs/2306.05685.
• Bai, Yuntao, et al. 2022. "Constitutional AI: Harmlessness from AI Feedback." arXiv:2212.08073.
• Vaswani, Ashish, et al. 2017. "Attention Is All You Need." In Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS’17), 6000–6010. Red Hook, NY: Curran Associates Inc.
• Kairouz, Peter, et al. 2021. "Advances and Open Problems in Federated Learning." March. http://arxiv.org/abs/1912.04977.
• McMahan, H. Brendan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Arcas. 2023. "Communication-Efficient Learning of Deep Networks from Decentralized Data." January. http://arxiv.org/abs/1602.05629.
• Abadi, Martin, et al. 2016. "Deep Learning with Differential Privacy." October. doi:10.1145/2976749.2978318.
• Tabassi, Elham. 2023. "Artificial Intelligence Risk Management Framework (AI RMF 1.0)." NIST. doi:10.6028/NIST.AI.100-1.