Jailbreaking Safeguarded Text-to-Image Models via Large Language Models
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866909983919046656 |
|---|---|
| author | Jiang, Zhengyuan Hu, Yuepeng Yang, Yuchen Cao, Yinzhi Gong, Neil Zhenqiang |
| author_facet | Jiang, Zhengyuan Hu, Yuepeng Yang, Yuchen Cao, Yinzhi Gong, Neil Zhenqiang |
| contents | Text-to-Image models may generate harmful content, such as pornographic images, particularly when unsafe prompts are submitted. To address this issue, safety filters are often added on top of text-to-image models, or the models themselves are aligned to reduce harmful outputs. However, these defenses remain vulnerable when an attacker strategically designs adversarial prompts to bypass these safety guardrails. In this work, we propose \alg, a method to jailbreak text-to-image models with safety guardrails using a fine-tuned large language model. Unlike other query-based jailbreak attacks that require repeated queries to the target model, our attack generates adversarial prompts efficiently after fine-tuning our AttackLLM. We evaluate our method on three datasets of unsafe prompts and against five safety guardrails. Our results demonstrate that our approach effectively bypasses safety guardrails, outperforms existing no-box attacks, and also facilitates other query-based attacks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_01839 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Jailbreaking Safeguarded Text-to-Image Models via Large Language Models Jiang, Zhengyuan Hu, Yuepeng Yang, Yuchen Cao, Yinzhi Gong, Neil Zhenqiang Cryptography and Security Artificial Intelligence Computation and Language Computer Vision and Pattern Recognition Text-to-Image models may generate harmful content, such as pornographic images, particularly when unsafe prompts are submitted. To address this issue, safety filters are often added on top of text-to-image models, or the models themselves are aligned to reduce harmful outputs. However, these defenses remain vulnerable when an attacker strategically designs adversarial prompts to bypass these safety guardrails. In this work, we propose \alg, a method to jailbreak text-to-image models with safety guardrails using a fine-tuned large language model. Unlike other query-based jailbreak attacks that require repeated queries to the target model, our attack generates adversarial prompts efficiently after fine-tuning our AttackLLM. We evaluate our method on three datasets of unsafe prompts and against five safety guardrails. Our results demonstrate that our approach effectively bypasses safety guardrails, outperforms existing no-box attacks, and also facilitates other query-based attacks. |
| title | Jailbreaking Safeguarded Text-to-Image Models via Large Language Models |
| topic | Cryptography and Security Artificial Intelligence Computation and Language Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2503.01839 |