EvoShield: Selective Test-Time Adaptation for Prompt Injection Detection via Active LLM Querying

Abstract

Prompt injection detection is commonly studied as a static offline classification problem, yet deployed LLM systems face evolving attacks and distribution shift after deployment. Static detectors are therefore poorly matched to the threat model, while routing every input to a stronger external LLM is costly and defeats the purpose of a local detector. We formulate prompt injection detection as a selective test-time adaptation problem. Our framework combines a prompt-based local detector built on

Categories

Framework mappings

Suggested from the entry's categories.

@article{zheng2026evoshield,
  title = {{EvoShield: Selective Test-Time Adaptation for Prompt Injection Detection via Active LLM Querying}},
  author = {Zanhong Zheng and Jieming Liang and Mengqin Hu and Yijuan Pei and Guobao Xu and Zhenlu Wu},
  year = {2026},
  month = may,
  journal = {Mathematics},
  doi = {10.3390/math14101719},
  url = {https://doi.org/10.3390/math14101719}
}