all InfoSec news
MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance
June 18, 2024, 4:19 a.m. | Renjie Pi, Tianyang Han, Jianshu Zhang, Yueqi Xie, Rui Pan, Qing Lian, Hanze Dong, Jipeng Zhang, Tong Zhang
cs.CR updates on arXiv.org arxiv.org
Abstract: The deployment of multimodal large language models (MLLMs) has brought forth a unique vulnerability: susceptibility to malicious attacks through visual inputs. This paper investigates the novel challenge of defending MLLMs against such attacks. Compared to large language models (LLMs), MLLMs include an additional image modality. We discover that images act as a ``foreign language" that is not considered during safety alignment, making MLLMs more prone to producing harmful responses. Unfortunately, unlike the discrete tokens considered …
arxiv attacks challenge cs.cl cs.cr cs.cv defending deployment discover image inputs language language models large llms malicious mllms multimodal novel performance safety vulnerability
More from arxiv.org / cs.CR updates on arXiv.org
Jobs in InfoSec / Cybersecurity
Information Technology Specialist I: Windows Engineer
@ Los Angeles County Employees Retirement Association (LACERA) | Pasadena, California
Information Technology Specialist I, LACERA: Information Security Engineer
@ Los Angeles County Employees Retirement Association (LACERA) | Pasadena, CA
Vice President, Controls Design & Development-7
@ State Street | Quincy, Massachusetts
Vice President, Controls Design & Development-5
@ State Street | Quincy, Massachusetts
Data Scientist & AI Prompt Engineer
@ Varonis | Israel
Contractor
@ Birlasoft | INDIA - MUMBAI - BIRLASOFT OFFICE, IN