Query-Based Adversarial Prompt Generation

arXiv:2402.12329v1 Announce Type: cross
Abstract: Recent work has shown it is possible to construct adversarial examples that cause an aligned language model to emit harmful strings or perform harmful behavior. Existing attacks work either in the white-box setting (with full access to the model weights), or through transferability: the phenomenon that adversarial examples crafted on one model often remain effective on other models. We improve on prior work with a query-based attack that leverages API access to a remote language …

abstract adversarial adversarial examples arxiv attacks behavior box construct cs.ai cs.cl cs.cr cs.lg examples language language model prompt query strings through type work