Large Vision-Language Models | Gelei Denghttps://geleideng.github.io/tags/large-vision-language-models/Large Vision-Language ModelsHugo Blox Builder (https://hugoblox.com)en-usTue, 02 Dec 2025 00:00:00 +0000https://geleideng.github.io/media/icon_hu7729264130191091259.pngLarge Vision-Language Modelshttps://geleideng.github.io/tags/large-vision-language-models/Safe + Safe = Unsafe? Exploring How Safe Images Can Be Exploited to Jailbreak Large Vision-Language Modelshttps://geleideng.github.io/publication/safe-safe-unsafe/Tue, 02 Dec 2025 00:00:00 +0000https://geleideng.github.io/publication/safe-safe-unsafe/<p>This paper identifies a multimodal safety failure mode where safe visual inputs can snowball into unsafe model behavior when combined with additional safe images and prompts. Safety Snowball Agent operationalizes this observation as a tool-using jailbreak framework for evaluating LVLM guardrails.</p>Safety Snowball Agenthttps://geleideng.github.io/project/safety-snowball-agent/Tue, 02 Dec 2025 00:00:00 +0000https://geleideng.github.io/project/safety-snowball-agent/<p>Safety Snowball Agent is an agent-based framework for evaluating how safe visual inputs can combine into unsafe behavior in large vision-language models.</p> <p>The framework accompanies the NeurIPS 2025 paper &ldquo;Safe + Safe = Unsafe?&rdquo; and probes a multimodal jailbreak mechanism that differs from traditional adversarial-image attacks.</p>