A Multi-Layered Detection and Prevention Framework for Prompt Injection Attacks in AI Systems

This research presents a robust framework for detecting and preventing prompt injection attacks in AI systems. By integrating multiple security layers, the system enhances reliability and trustworthiness. The results demonstrate that multi-layered defence significantly improves protection compared to traditional methods.

[1] F. Perez and I. Ribeiro, “Prompt Injection Attacks Against LLM Applications,” arXiv, 2023. [2] OWASP, “Top 10 Risks for LLM Applications,” 2024. [3] D. Kang et al., “Adversarial Attacks on LLMs,” IEEE Access, 2024. [4] X. Chen et al., “Security in NLP Systems,” IEEE, 2024. [5] OpenAI, “GPT-4 Technical Report,” 2023. [6] Y. Liu et al., “Jailbreaking LLMs,” arXiv, 2023. [7] S. Shen et al., “Prompt Engineering Security,” ACM, 2024. [8] A. Zou et al., “Adversarial Attacks on Language Models,” arXiv, 2023. [9] N. Carlini et al., “Extracting Training Data from LLMs,” USENIX, 2023. [10] J. Wei et al., “Emergent Abilities of LLMs,” arXiv, 2022.