Sparse Tokens Suffice: Jailbreaking Audio Language Models via Token-Aware Gradient Optimization

ArXi:2605.04700v1 Announce Type: cross Jailbreak attacks on audio language models (ALMs) optimize audio perturbations to elicit unsafe generations, and they typically update the entire waveform densely throughout optimization. In this work, we investigate the necessity of such dense optimization by analyzing the structure of token-aligned gradients in ALMs. We find that gradient energy is highly non-uniform across audio tokens, indicating that only a small subset of token-aligned audio regions dominates the optimization signal.