Fast Adversarial Attacks on Language Models In One GPU Minute

arXiv:2402.15570v1 Announce Type: new
Abstract: In this paper, we introduce a novel class of fast, beam search-based adversarial attack (BEAST) for Language Models (LMs). BEAST employs interpretable parameters, enabling attackers to balance between attack speed, success rate, and the readability of adversarial prompts. The computational efficiency of BEAST facilitates us to investigate its applications on LMs for jailbreaking, eliciting hallucinations, and privacy attacks. Our gradient-free targeted attack can jailbreak aligned LMs with high attack success rates within one minute. For …

adversarial adversarial attacks arxiv attacks cs.ai cs.cl cs.cr fast gpu language language models