SPIRIT: Patching Speech Language Models against Jailbreak Attacks

2025-05-18 · Amirbek Djanibekov, Nurdaulet Mukhituly, Kentaro Inui, Hanan Aldarmaki, Nils Lukas

Speech Language Models (SLMs) enable natural interactions via spoken instructions, which more effectively capture user intent by detecting nuances in speech. The richer speech signal introduces new security risks compared to text-based models, as adversaries can better bypass safety mechanisms by injecting imperceptible noise to speech. We analyze adversarial attacks and find that SLMs are substantially more vulnerable to jailbreak attacks, which can achieve a perfect 100% attack success rate in some instances. To improve security, we propose post-hoc patching defenses used to intervene during inference by modifying the SLM's activations that improve robustness up to 99% with (i) negligible impact on utility and (ii) without any re-training. We conduct ablation studies to maximize the efficacy of our defenses and improve the utility/security trade-off, validated with large-scale benchmarks unique to SLMs.

📄 PDF Abstract BibTeX arXiv:2505.13541

Code (0)

등록된 구현이 없습니다.

Methods 이 논문이 사용한 방법론

Patching Activation patching studies the model's computation by altering its latent representations, the token embeddings in transformer-based language models, during the inference process

Attributing and Exploiting Safety Vectors through Global Optimization in Large Language Models

2026-01-22 · Fengheng Chu, Jiahao Chen, Yuhong Wang, Jun Wang 외 arxiv

While Large Language Models (LLMs) are aligned to mitigate risks, their safety guardrails remain fragile against jailbreak attacks. This reveals limited understanding of components governing safety. Existing methods rely…

Spirit LM: Interleaved Spoken and Written Language Model

2024-02-08 · Tu Anh Nguyen, Benjamin Muller, Bokai Yu, Marta R. Costa-Jussa 외

We introduce Spirit LM, a foundation multimodal language model that freely mixes text and speech. Our model is based on a 7B pretrained text language model that we extend to the speech modality by continuously training i…

Language Modeling Language Modelling model

AudioJailbreak: Jailbreak Attacks against End-to-End Large Audio-Language Models

2025-05-20 · Guangke Chen, Fu Song, Zhe Zhao, Xiaojun Jia 외

Jailbreak attacks to Large audio-language models (LALMs) are studied recently, but they achieve suboptimal effectiveness, applicability, and practicability, particularly, assuming that the adversary can fully manipulate …

text-to-speech Text to Speech

Patcher: Post-Hoc Patching of Backdoored Large Language Models

2026-06-02 · Anjun Gao, Yueyang Quan, Yufei Xia, Zhuqing Liu 외 arxiv

Large language models remain vulnerable to jailbreak backdoor attacks, where adversaries poison safety alignment data to embed hidden triggers that bypass safety mechanisms. Existing defenses often require comprehensive …

Sirens' Whisper: Inaudible Near-Ultrasonic Jailbreaks of Speech-Driven LLMs

2026-03-14 · Zijian Ling, Pingyi Hu, Xiuyong Gao, Xiaojing Ma 외 arxiv

Speech-driven large language models (LLMs) are increasingly accessed through speech interfaces, introducing new security risks via open acoustic channels. We present Sirens' Whisper (SWhisper), the first practical framew…