PromptSleuth: Detecting Prompt Injection via Semantic Intent Invariance

A detector for prompt-injection attacks that reasons about the semantic intent of the input rather than matching surface patterns. Intent-invariant checks stay robust when attackers paraphrase, encode, or reorder the injected instruction.

Recommended citation: Mengxiao Wang, Yuxuan Zhang, and Guofei Gu. (2025). "PromptSleuth: Detecting Prompt Injection via Semantic Intent Invariance." arXiv preprint arXiv:2508.20890.
Download Paper