We present the Korean Multimodal INteraction Data (K-MIND), a large-scale corpus of dyadic Korean dialogue that is designed to capture the multimodal richness of social interaction. The dataset includes 292 participants and 200 sets (935 clips) spanning 115 hours and 30 minutes, all aligned across verbal, paraverbal, and nonverbal modalities such as transcripts, acoustic features, and visual signals. For these modalities, we propose a comprehensive annotation scheme that enables nuanced yet consistent labeling of complex communicative behaviors, balancing theoretical soundness with practical feasibility. We further report analysis results of the corpus, including label distributions, within- and cross-layer analyses. These analyses illuminate the key properties of dyadic K-MIND and demonstrate its utility for advancing research in human–computer interaction as well as in interdisciplinary domains. To ensure continuous refinement, the corpus and framework are being validated in complementary studies and have been extended to triadic interactions (K-MIND Triadic) that model group dynamics, which will be included in upcoming releases.

Large language model (LLM) safety benchmarks often evaluate input risk, response selection, generation, and tool use separately, making comparisons across tasks difficult. We introduce KoEVD (Korean Ethical Violation Dataset), which links five evaluation and analysis tasks through annotated Korean source utterances: input-risk judgment, candidate-response selection, generation safety, response-strategy analysis, and non-executable mock-tool selection. The tasks share source IDs, with task-specific subsets and candidate sampling where required. Built through an approximately 3.5-year human-governed, AI-assisted process, KoEVD contains 13,552 curated user utterances and 71,395 linked response candidates. Across 12 open-weight and commercial models, observed task profiles differ, and aggregate scores obscure variation across severity, explicitness, utterance type, and topic. In matched generation and mock-tool-selection cases, responses classified as safe coexist with unsafe mock-tool choices; pooled unsafe-call rates are 9.73% and 9.31% for the classifier-safe and classifier-unsafe groups, respectively, with framing-specific differences. These diagnostic measurements do not establish real-world safety. KoEVD supports detailed Korean safety analysis through contextual annotations and source-linked comparisons.