PVDetector: Detecting Prompt Injection Attacks on Purpose-Specific LLM Agents through Policy-Violation Concept Analysis

Part of

Adaptive Interfaces And AI Benchmarking