Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2509.12521 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912588811468800 |
|---|---|
| author | Lan, Yifan Cao, Yuanpu Zhang, Weitong Lin, Lu Chen, Jinghui |
| author_facet | Lan, Yifan Cao, Yuanpu Zhang, Weitong Lin, Lu Chen, Jinghui |
| contents | Recently, Multimodal Large Language Models (MLLMs) have gained significant attention across various domains. However, their widespread adoption has also raised serious safety concerns. In this paper, we uncover a new safety risk of MLLMs: the output preference of MLLMs can be arbitrarily manipulated by carefully optimized images. Such attacks often generate contextually relevant yet biased responses that are neither overtly harmful nor unethical, making them difficult to detect. Specifically, we introduce a novel method, Preference Hijacking (Phi), for manipulating the MLLM response preferences using a preference hijacked image. Our method works at inference time and requires no model modifications. Additionally, we introduce a universal hijacking perturbation -- a transferable component that can be embedded into different images to hijack MLLM responses toward any attacker-specified preferences. Experimental results across various tasks demonstrate the effectiveness of our approach. The code for Phi is accessible at https://github.com/Yifan-Lan/Phi. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_12521 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Phi: Preference Hijacking in Multi-modal Large Language Models at Inference Time Lan, Yifan Cao, Yuanpu Zhang, Weitong Lin, Lu Chen, Jinghui Machine Learning Recently, Multimodal Large Language Models (MLLMs) have gained significant attention across various domains. However, their widespread adoption has also raised serious safety concerns. In this paper, we uncover a new safety risk of MLLMs: the output preference of MLLMs can be arbitrarily manipulated by carefully optimized images. Such attacks often generate contextually relevant yet biased responses that are neither overtly harmful nor unethical, making them difficult to detect. Specifically, we introduce a novel method, Preference Hijacking (Phi), for manipulating the MLLM response preferences using a preference hijacked image. Our method works at inference time and requires no model modifications. Additionally, we introduce a universal hijacking perturbation -- a transferable component that can be embedded into different images to hijack MLLM responses toward any attacker-specified preferences. Experimental results across various tasks demonstrate the effectiveness of our approach. The code for Phi is accessible at https://github.com/Yifan-Lan/Phi. |
| title | Phi: Preference Hijacking in Multi-modal Large Language Models at Inference Time |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2509.12521 |