When Good Sounds Go Adversarial: Jailbreaking Audio-Language Models with Benign Inputs

Open in new window