Data Breach Prevention in AI Systems: Employing Event-Driven Architecture to Combat Prompt Injection Attacks in Chatbots
Mary Jane C. Samonte, Jon Eril R. Aparize, Ethan Joseph S. Gonzales, Joyful L. Morilla · 2024
While incorporating Large Language Models (LLMs) into applications has transformed user experiences, it has also brought about serious security flaws, most notably quick injection attacks. These attacks risk user privacy and data integrity by manipulating system behavior with AI model defects. This study thoroughly reviews the literature on quick injection attacks in chatbots to examine vulnerabilities in AI system integration. It suggests different strategies and synthesizes them into an event-driven architecture (EDA) that is specifically designed to thwart rapid injection assaults by synthesizing the findings of previous research. The suggested EDA offers a scientific breakthrough in comprehending and resolving security challenges in AI system integration by enabling proactive threat identification and mitigation. The results highlight how crucial strong architectural security and systems integration is to defend AI systems from changing threats. The importance of these actions in strengthening cybersecurity standards and safeguarding digital environments is also covered in the report. This study adds to the current conversation around AI system security by offering ideas and viewpoints for more investigation and advancement.