# Werewolf benchmark shows GPT models miss social cues in deception

> Original title: EMNLP’26｜GPT也不会「读空气」，狼人杀基准测出社交盲区

- Source: AI Era
- Published: 2026-10-29T00:00:22.000Z
- Bad Influence: https://badinfluence.news/items/ox0ygz2mlu0x67dolqykigf30
- Original: https://aiera.com.cn/2026/10/29/other/admin/121106/emnlp26%ef%bd%9cgpt%e4%b9%9f%e4%b8%8d%e4%bc%9a%e3%80%8c%e8%af%bb%e7%a9%ba%e6%b0%94%e3%80%8d%ef%bc%8c%e7%8b%bc%e4%ba%ba%e6%9d%80%e5%9f%ba%e5%87%86%e6%b5%8b%e5%87%ba%e7%a4%be%e4%ba%a4%e7%9b%b2/

## Summary

Researchers from Dongxi NLP and collaborators built a Werewolf game benchmark to test whether multimodal large models can detect deception from video, audio, and text. The source text provided is only a short lead, so no benchmark scores, model versions, or findings are confirmed here.