Reproduce “DailySafe: Benchmarking Dynamic Safety Awareness in Everyday Multi-Turn Multimodal Interactions” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xuankun Rong, Xinhao Deng, Wenzhang Jiang, Ziheng Peng, Siyuan Liang, Wenke Huang, Pablo Arbelaez, Mang Ye

DailySafe evaluates whether multimodal assistants can track changing risks across everyday tasks, using 195 interactions with evidence distributed across turns and images. It measures safety and helpfulness separately across 16 models. Claude Opus 5.5 achieved the highest mean safety rate, 92.65%, and tied GPT-6 Sol for highest helpfulness, 94.53%. These are benchmark results, not real-world reliability estimates.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.