Embedding-based classifiers can detect prompt injection attacks

[Free GPT-4]
[PDF] arxiv.org

Gandalf the Red: Adaptive Security for LLMs

N Pfister, V Volhejn, M Knott, S Arias… - arxiv preprint arxiv …, 2025 - arxiv.org
Current evaluations of defenses against prompt attacks in large language model (LLM)
applications often overlook two critical factors: the dynamic nature of adversarial behavior …
Uložit Citovat Související články Všechny verze (počet: 2) Zobrazit jako HTML