prompt-level attack

token-level attack (1) large language model (1) defense method (1) security evaluation (1) zero-shot manner (1) jailbreak attack (1) ai safety (1) model safety (1) adversarial learning (1) benchmark evaluation (1)