{"ID":2824520,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2512.23881","arxiv_id":"2512.23881","title":"Breaking Audio Large Language Models by Attacking Only the Encoder: A Universal Targeted Latent-Space Audio Attack","abstract":"Audio-language models combine audio encoders with large language models to enable multimodal reasoning, but they also introduce new security vulnerabilities. We propose a universal targeted latent space attack, an encoder-level adversarial attack that manipulates audio latent representations to induce attacker-specified outputs in downstream language generation. Unlike prior waveform-level or input-specific attacks, our approach learns a universal perturbation that generalizes across inputs and speakers and does not require access to the language model. Experiments on Qwen2-Audio-7B-Instruct demonstrate consistently high attack success rates with minimal perceptual distortion, revealing a critical and previously underexplored attack surface at the encoder level of multimodal systems.","short_abstract":"Audio-language models combine audio encoders with large language models to enable multimodal reasoning, but they also introduce new security vulnerabilities. We propose a universal targeted latent space attack, an encoder-level adversarial attack that manipulates audio latent representations to induce attacker-specifie...","url_abs":"https://arxiv.org/abs/2512.23881","url_pdf":"https://arxiv.org/pdf/2512.23881v1","authors":"[\"Roee Ziv\",\"Raz Lapid\",\"Moshe Sipper\"]","published":"2025-12-29T21:56:13Z","proceeding":"cs.SD","tasks":"[\"cs.SD\",\"cs.AI\",\"cs.CR\"]","methods":"[\"Language Model\"]","has_code":false}