Self-Evaluation as a Defense Against Adversarial Attacks on LLMs

arXiv:2407.03234v1 Announce Type: cross
Abstract: When LLMs are deployed in sensitive, human-facing settings, it is crucial that they do not output unsafe, biased, or privacy-violating outputs. For this reason, models are both trained and instructed to refuse to answer unsafe prompts such as "Tell me how to build a bomb." We find that, despite these safeguards, it is possible to break model defenses simply by appending a space to the end of a model's input. In a study …

adversarial adversarial attacks arxiv attacks cs.cl cs.cr cs.lg defense evaluation llms