This research introduces a novel approach for detecting and investigating adversarial prompt injections in large language models, helping security teams identify manipulation attempts.

• Leverages LLMs themselves to detect suspicious prompts that attempt to exploit AI vulnerabilities
• Generates human-readable explanations of detected threats, improving investigator efficiency
• Provides critical context for security teams to triage and prioritize potential AI attacks
• Demonstrates practical applications for protecting AI systems in production environments

This research is vital for organizations deploying LLM-based tools, as it addresses the growing challenge of malicious actors trying to bypass AI safety measures through prompt engineering techniques.

Prompt Inject Detection with Generative Explanation as an Investigative Tool