Mitigation of Data Leaks and Prompt Injection in University Chatbots Using a Decoupled Perimeter Architecture
DOI:
https://doi.org/10.62452/40j21f47Keywords:
Large Language Models, Prompt Inject, data leakage, artificial intelligence, higher education , cybersecurity competenciesAbstract
Higher education is rapidly adopting Generative Artificial Intelligence tools, a situation that has exposed universities to critical cybersecurity vulnerabilities, mainly prompt injection and data leakage. This study designs and evaluates an architecture for a decentralized conversational agent that functions as a virtual course assistant, taking into account the cybersecurity recommendations of the faculty. The methodological approach is theoretical-experimental, and the research is conducted in three phases: an assessment of faculty members’ cybersecurity knowledge via a digital questionnaire; the design of perimeter middleware that mitigates data leakage and is effective against prompt injection; and the experimental evaluation of the design. The experimental results demonstrate that the proposed architectural design significantly reduces data exposure and vulnerability to prompt injection attacks. The research provides a practical and replicable solution that strengthens privacy, information security, and the responsible adoption of AI in the university setting.
Downloads
References
Alafnan, M. A., Dishari, S., Jovic, M., & Lomidze, K. (2023). ChatGPT as an educational tool: Opportunities, challenges, and recommendations for communication, business writing, and composition courses. Journal of Artificial Intelligence and Technology, 3(2), 60–68. https://doi.org/10.37965/jait.2023.0184
Anghel, C., Craciun, M. V., Cocu, A., Anghel, A. A., & Istrate, A. (2026). SteadyEval: Robust LLM exam graders via adversarial training and distillation. Computers, 15(1). https://doi.org/10.3390/computers15010055
Chroma Core Inc. (2024). The open-source embedding database (Version 0.5.x) [Software]. https://www.trychroma.com/
Cotton, D. R. E., Cotton, P. A., & Shipway, J. R. (2024). Chatting and cheating: Ensuring academic integrity in the era of ChatGPT. Innovations in Education and Teaching International, 61(2), 228–239. https://doi.org/10.1080/14703297.2023.2190148
De Freitas, J., & Cohen, I. G. (2025). Disclosure, humanizing, and contextual vulnerability of generative AI chatbots. NEJM AI, 2(2). https://doi.org/10.1056/aipc2400464
Dzhaliuk, N., Sabodashko, D., Khoma, V., Khoma, Y., Kolchenko, V., & Podpora, M. (2026). Comparative evaluation of machine learning methods for protecting LLMs from prompt injection attacks. International Journal of Information Security, 25. https://doi.org/10.1007/s10207-026-01264-8
Gamez Lagos, N. M. (2025). Asistente virtual académico basado en sistemas LLM para procesar y extraer información normativa de instituciones de educación superior. https://www.researchgate.net/publication/395710153_Asistente_Virtual_Academico_Basado_en_Sistemas_LLM_para_Procesar_y_Extraer_Informacion_Normativa_de_Instituciones_de_Educacion_Superior
Gayathri, T., Durusoju, S., Mergoju, L., & Reddy, P. V. (2025). MedicAI: AI-powered system for enhanced medical information access. 9th International Conference on Inventive Systems and Control (ICISC). Coimbatore, India.
Giapantzis, K., Bountakas, P., Zarras, A., Farao, A., Bolgouras, V., & Xenakis, C. (2026). Adaptive DeSeTra: Adaptive deformable-span self-attention transformer for LLM security. Information Sciences, 754. https://doi.org/10.1016/j.ins.2026.123670
Gundu, T. (2023). Chatbots: A framework for improving information security behaviours using ChatGPT. En S. Furnell & N. Clarke (Eds.), Human aspects of information security and assurance: HAISA 2023 (Vol. 674, pp. 418–431). Springer.
Johno, H., Amakawa, A., Komaba, A., Tozuka, R., Johno, Y., Sato, J., Yoshimura, K., Nakamoto, K., & Ichikawa, S. (2026). Open-source offline-deployable retrieval-augmented large language model for assisting pancreatic cancer staging. Japanese Journal of Radiology. https://doi.org/10.1007/s11604-026-02049-8
Kumar, S. S., Cummings, M. L., & Stimpson, A. (2024). Strengthening LLM trust boundaries: A survey of prompt injection attacks. 4th International Conference on Human-Machine Systems (ICHMS). Toronto, Canada.
Li, Y., Fu, X., Verma, G., Buitelaar, P., & Liu, M. (2025). Mitigating hallucination in large language models (LLMs): An application-oriented survey on RAG, reasoning, and agentic systems. arXiv. http://arxiv.org/abs/2510.24476
Luo, H., & Ji, C. (2025). Federated learning-based data collaboration method for enhancing edge cloud AI system security using large language models. 5th International Symposium on Computer Technology and Information Science (ISCTIS 2025) (pp. 163–166). IEEE. Xi'an, China.
Microsoft. (2023). Presidio: Data protection and anonymization SDK [Software]. Microsoft Open Source. https://presidio.dataprivacystack.org/
Ollama. (2023). Get up and running with Llama 3 and other large language models locally [Software]. https://ollama.com/
Rebuff AI. (2023). Rebuff: Prompt injection detection service for LLM applications [Software]. GitHub. https://github.com/rebuff-ai/rebuff
Saad, M., Mughal, S. F., Mustafa, M., Ahmed, I., & Masood, A. (2025). Assessing and mitigating multi-turn jailbreak vulnerabilities in large language models: A Crescendo attack study. 27th International Multitopic Conference (INMIC 2025). Islamabad, Pakistan.
Streamlit. (2026). Streamlit Community Cloud. Snowflake Inc. https://share.streamlit.io/
Toro, Y., Toro, I., Calvo, C., Urquizo, D., & Chapalbay, M. (2025). Implementación de un asistente virtual de inteligencia artificial en universidades latinoamericanas. Invecom, 6(3). https://revistainvecom.org/index.php/invecom/article/view/4258/4885
Trinks, M., Gondim, J., & Albuquerque, R. (2023). Multi-agent architecture for passive rootkit detection with data enrichment. In M. V. Garcia & C. Gordón-Gallegos (Eds.), CSEI: International Conference on Computer Science, Electronics and Industrial Engineering (CSEI 2022) (Vol. 678, pp. 29–41). Springer.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Jorge Wilson Saa-Saltos, Nancy Magali Rodríguez-Gavilanes (Autor/a)
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
Authors who publish in Revista Metropolitana de Ciencias Aplicadas (REMCA), agree to the following terms:
1. Copyright
Authors retain unrestricted copyright to their work. Authors grant the journal the right of first publication. To this end, they assign the journal non-exclusive exploitation rights (reproduction, distribution, public communication, and transformation). Authors may enter into additional agreements for the non-exclusive distribution of the version of the work published in the journal, provided that acknowledgment of its initial publication in this journal is given.
© The authors.
2. License
The articles are published in the journal under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0). The terms can be found at: https://creativecommons.org/licenses/by-nc-sa/4.0/deed.en
This license allows:
- Sharing: Copying and redistributing the material in any medium or format.
- Adapting: Remixing, transforming, and building upon the material.
Under the following terms:
- Attribution: You must give appropriate credit, provide a link to the license, and indicate if any changes were made. You may do this in any reasonable manner, but not in any way that suggests the licensor endorses or sponsors your use.
- NonCommercial: You may not use the material for commercial purposes.
- ShareAlike: If you remix, transform, or build upon the material, you must distribute your creation under the same license as the original work.
There are no additional restrictions. You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits.
