Self-Aware Safety Augmentation: Leveraging Internal Semantic Understanding to Enhance Safety in Vision-Language Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909710517534720 |
|---|---|
| author | Wang, Wanying Ma, Zeyu Zheng, Han Tan, Xin Chen, Mingang |
| author_facet | Wang, Wanying Ma, Zeyu Zheng, Han Tan, Xin Chen, Mingang |
| contents | Large vision-language models (LVLMs) are vulnerable to harmful input compared to their language-only backbones. We investigated this vulnerability by exploring LVLMs internal dynamics, framing their inherent safety understanding in terms of three key capabilities. Specifically, we define these capabilities as safety perception, semantic understanding, and alignment for linguistic expression, and experimentally pinpointed their primary locations within the model architecture. The results indicate that safety perception often emerges before comprehensive semantic understanding, leading to the reduction in safety. Motivated by these findings, we propose \textbf{Self-Aware Safety Augmentation (SASA)}, a technique that projects informative semantic representations from intermediate layers onto earlier safety-oriented layers. This approach leverages the model's inherent semantic understanding to enhance safety recognition without fine-tuning. Then, we employ linear probing to articulate the model's internal semantic comprehension to detect the risk before the generation process. Extensive experiments on various datasets and tasks demonstrate that SASA significantly improves the safety of LVLMs, with minimal impact on the utility. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_21637 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Self-Aware Safety Augmentation: Leveraging Internal Semantic Understanding to Enhance Safety in Vision-Language Models Wang, Wanying Ma, Zeyu Zheng, Han Tan, Xin Chen, Mingang Artificial Intelligence Large vision-language models (LVLMs) are vulnerable to harmful input compared to their language-only backbones. We investigated this vulnerability by exploring LVLMs internal dynamics, framing their inherent safety understanding in terms of three key capabilities. Specifically, we define these capabilities as safety perception, semantic understanding, and alignment for linguistic expression, and experimentally pinpointed their primary locations within the model architecture. The results indicate that safety perception often emerges before comprehensive semantic understanding, leading to the reduction in safety. Motivated by these findings, we propose \textbf{Self-Aware Safety Augmentation (SASA)}, a technique that projects informative semantic representations from intermediate layers onto earlier safety-oriented layers. This approach leverages the model's inherent semantic understanding to enhance safety recognition without fine-tuning. Then, we employ linear probing to articulate the model's internal semantic comprehension to detect the risk before the generation process. Extensive experiments on various datasets and tasks demonstrate that SASA significantly improves the safety of LVLMs, with minimal impact on the utility. |
| title | Self-Aware Safety Augmentation: Leveraging Internal Semantic Understanding to Enhance Safety in Vision-Language Models |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2507.21637 |