Jailbreaking Large Vision Language Models in Intelligent Transportation Systems
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866918206877204480 |
|---|---|
| author | Das, Badhan Chandra Jawad, Md Tasnim Mia, Md Jueal Amini, M. Hadi Wu, Yanzhao |
| author_facet | Das, Badhan Chandra Jawad, Md Tasnim Mia, Md Jueal Amini, M. Hadi Wu, Yanzhao |
| contents | Large Vision Language Models (LVLMs) demonstrate strong capabilities in multimodal reasoning and many real-world applications, such as visual question answering. However, LVLMs are highly vulnerable to jailbreaking attacks. This paper systematically analyzes the vulnerabilities of LVLMs integrated in Intelligent Transportation Systems (ITS) under carefully crafted jailbreaking attacks. First, we carefully construct a dataset with harmful queries relevant to transportation, following OpenAI's prohibited categories to which the LVLMs should not respond. Second, we introduce a novel jailbreaking attack that exploits the vulnerabilities of LVLMs through image typography manipulation and multi-turn prompting. Third, we propose a multi-layered response filtering defense technique to prevent the model from generating inappropriate responses. We perform extensive experiments with the proposed attack and defense on the state-of-the-art LVLMs (both open-source and closed-source). To evaluate the attack method and defense technique, we use GPT-4's judgment to determine the toxicity score of the generated responses, as well as manual verification. Further, we compare our proposed jailbreaking method with existing jailbreaking techniques and highlight severe security risks involved with jailbreaking attacks with image typography manipulation and multi-turn prompting in the LVLMs integrated in ITS. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_13892 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Jailbreaking Large Vision Language Models in Intelligent Transportation Systems Das, Badhan Chandra Jawad, Md Tasnim Mia, Md Jueal Amini, M. Hadi Wu, Yanzhao Artificial Intelligence Large Vision Language Models (LVLMs) demonstrate strong capabilities in multimodal reasoning and many real-world applications, such as visual question answering. However, LVLMs are highly vulnerable to jailbreaking attacks. This paper systematically analyzes the vulnerabilities of LVLMs integrated in Intelligent Transportation Systems (ITS) under carefully crafted jailbreaking attacks. First, we carefully construct a dataset with harmful queries relevant to transportation, following OpenAI's prohibited categories to which the LVLMs should not respond. Second, we introduce a novel jailbreaking attack that exploits the vulnerabilities of LVLMs through image typography manipulation and multi-turn prompting. Third, we propose a multi-layered response filtering defense technique to prevent the model from generating inappropriate responses. We perform extensive experiments with the proposed attack and defense on the state-of-the-art LVLMs (both open-source and closed-source). To evaluate the attack method and defense technique, we use GPT-4's judgment to determine the toxicity score of the generated responses, as well as manual verification. Further, we compare our proposed jailbreaking method with existing jailbreaking techniques and highlight severe security risks involved with jailbreaking attacks with image typography manipulation and multi-turn prompting in the LVLMs integrated in ITS. |
| title | Jailbreaking Large Vision Language Models in Intelligent Transportation Systems |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2511.13892 |