Universal Vulnerabilities in Large Language Models: Backdoor Attacks for In-context Learning

arXiv:2401.05949v4 Announce Type: cross
Abstract: In-context learning, a paradigm bridging the gap between pre-training and fine-tuning, has demonstrated high efficacy in several NLP tasks, especially in few-shot settings. Despite being widely applied, in-context learning is vulnerable to malicious attacks. In this work, we raise security concerns regarding this paradigm. Our studies demonstrate that an attacker can manipulate the behavior of large language models by poisoning the demonstration context, without the need for fine-tuning the model. Specifically, we design a new …

arxiv attacks backdoor backdoor attacks bridging the gap context cs.ai cs.cl cs.cr fine-tuning gap high language language models large malicious nlp paradigm security security concerns settings training vulnerabilities vulnerable work