UniGuardian offers the first unified defense system against various prompt-based attacks on Large Language Models, protecting against manipulation that could generate harmful outputs.

• Introduces the concept of Prompt Trigger Attacks (PTA) that unifies prompt injection, backdoor, and adversarial attacks
• Employs a novel detection mechanism that identifies poisoned prompts by analyzing linguistic patterns and semantic inconsistencies
• Demonstrates effective protection against multiple attack vectors with a single defense system
• Provides a practical security solution applicable to real-world LLM deployments

This research is critical for organizations deploying LLMs in production environments, as it addresses significant security vulnerabilities that could otherwise lead to harmful outputs, data leakage, or system manipulation.

UniGuardian: A Unified Defense for Detecting Prompt Injection, Backdoor Attacks and Adversarial Attacks in Large Language Models