Real-world fact-checking faces the challenge of temporal dynamics, where misinformation manifests as evolving evidences rather than static claims. Verifying such claims necessitates the synthesis of multi-hop fragmented evidence dispersed across the timeline. However, existing benchmarks largely neglect this complexity, limiting evaluation to static settings that fail to capture real-world demands. To address this, we pioneer MTIMEFACTCHECK, the first dynamic benchmark to rigorously evaluate multi-hop temporal fact-checking in realistic, evolving contexts. Utilizing MTIMEFACTCHECK, we curate the MTimeFactCorpus dataset to benchmark sixteen LLM baselines, scrutinizing their capabilities and critical limitations. Building on these findings, we tentatively propose a novel method MTIMEFACTCHECK to augment LLM efficacy in multi-hop temporal fact-checking, offering guidance for future research. Our data and code are available at https://github.com/SCUNLP/MTimeFact.

The effectiveness of large language models (LLMs) to fact-check misinformation remains uncertain, despite their growing use. To this end, we present CANDY, a benchmark designed to systematically evaluate the capabilities and limitations of LLMs in fact-checking Chinese misinformation. Specifically, we curate a carefully annotated dataset of ~20k instances. Our analysis shows that current LLMs exhibit limitations in generating accurate fact-checking conclusions, even when enhanced with chain-of-thought reasoning and few-shot prompting. To understand these limitations, we develop a taxonomy to categorize flawed LLM-generated explanations for their conclusions and identify factual fabrication as the most common failure mode. Although LLMs alone are unreliable for fact-checking, our findings indicate their considerable potential to augment human performance when deployed as assistive tools in scenarios. Our dataset and code can be accessed at https://github.com/SCUNLP/CANDY.