Large language models (LLMs) are increasingly used to guide urban safety decisions, but this study shows that their judgments are more influenced by neighborhood names than by geographic coordinates. Across seven instruct‑tuned models tested on 186 neighborhoods in Los Angeles and Chicago, name‑based ratings varied significantly and correlated with the proportion of locally dominant marginalized groups, while coordinate‑only ratings remained largely flat. The research finds that removing neighborhood names reduces both bias and accuracy, highlighting the complex role of demographic stereotypes and crime signals in LLM safety assessments.
By Huy Nguyen, Yue Lin
arXiv:2606. 15890v1 Announce Type: new Abstract: Understanding urban wellbeing from multimodal data requires integrating heterogeneous spatial and temporal signals, posing significant challenges for current multimodal large language models (MLLMs).
By Yanxin Xi, Xiang Su, Jie Feng, Yu Liu, Sasu Tarkoma, Pan Hui
arXiv:2606. 00871v1 Announce Type: cross Abstract: Vision-language models (VLMs) are increasingly used to generate structured descriptions of street-level imagery for tasks such as streetscape auditing, mapping, and public consultation.
By Rashid Mushkani
The paper investigates how large language and visual‑language models used in autonomous vehicles inherit human driver biases when deciding whether to yield to pedestrians. It introduces two new bias‑testing methods—All Else Being Equal and Self‑Consistency tests—to evaluate these models. Results reveal that the models’ yielding decisions are influenced by pedestrian attributes such as gender, ethnicity, religion, disability, age, skin tone, and socio‑economic status, with varying patterns across models.
By Irem Yoldas, Martim Brand\~ao, Jie Zhang, Odinaldo Rodrigues
arXiv:2608. 20026v1 Announce Type: cross Abstract: Streetscape quality has become a central concern in contemporary urban planning, particularly within the framework of the pedestrian-friendly 15-minute city, where walkability and public-space quality are increasingly recognized as key determinants of urban performance.
By Joan Perez, Giovanni Fusco
The paper introduces MPS-Bench, a benchmark of 5,181 scenarios from 584 real-world images across 12 high-risk domains, each paired with a hidden user profile, to evaluate personalized safety in vision‑language models (VLMs). Eight leading VLMs were tested and found to almost always respond directly (86‑99%) without seeking missing context, scoring no higher than 2.6/5 on personalized safety. The authors identify a phenomenon called visual dominance, where visual information enters text representations early and suppresses textual risk signals, and propose PRISM, a lightweight input monitor that predicts when a query should be deferred, achieving 0.978 AUC and outperforming all tested models on the safety‑utility Pareto frontier.
By Edward Sun, Yuchen Wu, Zixian Ma, Eric Hanchen Jiang, Yijia Xiao, Xiaoyuan Yi, Ranjay Krishna, Wei Wang, Jindong Wang, Aylin Caliskan
The paper investigates cross‑modal safety drift in multimodal large language models, where a harmless text query paired with a visual image can trigger harmful responses. Empirical analysis identifies unsafe response patterns and shows that visual cues receive limited attention, weakening refusal mechanisms. The authors introduce Safety‑Awareness Representation Transfer (SRT), a lightweight method that transfers safety signals from text processing to mitigate cross‑modal drift while maintaining model utility.
By Tianqi Xiao, Shiyao Cui, Minghao Zhang, Junxiao Yang, Renmiao Chen
Public trust in Autonomous Vehicles (AVs) may depend not only on technical success but also on the fairness of their decision making. While a recent trend in AV research involves using general purpose...
arXiv:2606. 00369v1 Announce Type: cross Abstract: Safe global deployment of AI models requires alignment with human values that vary across cultures.
By Arkadiy Saakyan, Charvi Rastogi, Lora Aroyo
Vision‑language models used to gauge urban change from repeated street‑level images exhibit limited reliability at single locations. In a study of 4,648 image pairs from 435 Google Street View points across five U.S. cities, re‑photographing the same street altered perception scores by an average of 0.80 points—about two‑thirds of the difference between distinct streets—while repeated model calls added negligible variation. Although image re‑encoding, prompt order, and various image statistics contributed modestly, a small systematic drift (~0.1 points) persisted and grew with time between captures, suggesting minor unrecorded physical changes. Controlled experiments revealed that varying camera and image properties can shift scores, and that camera geometry alone caused a model to falsely report change in 45% of identical scenes; normalising to a common virtual camera reduced this to 7.5%. Despite these individual‑point unreliabilities, aggregating many paired observations recovers a clear redevelopment signal, indicating that such models are dependable at large scales but not for single‑location assessments.
By Kaizhen Tan
Assessing proxemic danger from a robot's egocentric perspective is critical for safe embodied navigation in human environments and requires both visual and contextual reasoning. We evaluate three opensource vision-language models (VLMs) (\textit{InternVL}, \textit{Qwen-VL}, and \textit{SmolVLM}) on the classification of egocentric robot images into four danger levels, comparing three prompting strategies and two rounds of QLoRA fine-tuning against a stratified random baseline.
arXiv:2609.20850v1 Announce Type: new
Abstract: While Multimodal Large Language Models (MLLMs) show remarkable advancements, their cross-modal capabilities introduce complex vulnerabilities that easi...
By Yueming Lyu, Yilian Shi, Haoxiang Tan, Linzhuang Zou, Qihao Wang, Guihua Yu, Jie Qin, Xin Gao, Chenyang Si, Jing Dong, Caifeng Shan