Today for AI
HOT RADAR
llmHEAT 5.9°

AI CURATED HOT STORY · 10/28/2026

UTokyo Releases MIDA Benchmark Exposing Multimodal LLM Social Deception Blind Spots

1 reports archived1 independent sourcesupdated 10/28/2026, 19:50:00
Synthesis & Latest Updates
1 Sources Cross-Validated

A team from the University of Tokyo introduced the MIDA benchmark for multimodal interactive deception assessment, accepted at EMNLP 2026. Tests reveal that leading multimodal LLMs like GPT and Claude struggle to reliably distinguish truth from lies in social interactions, despite processing audio-visual-text inputs. The study highlights significant gaps in models' ability to 'read the room' and understand theory of mind.

LATEST/A team from the University of Tokyo introduced the MIDA benchmark for multimodal interactive deception assessment, accepted at EMNLP 2026. Tests reveal that leading multimodal LLMs like GPT and Claude struggle to reliably distinguish truth from lies in social interactions, despite processing audio-visual-text inputs. The study highlights significant gaps in models' ability to 'read the room' and understand theory of mind.

Hourly heat curve

3 fully observed hours
Now
5.9
Peak
6.310-30
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first
  1. 新智元 (微信公众号)T2·72 pts

    UTokyo Releases MIDA Benchmark Exposing Multimodal LLM Social Deception Blind Spots

    • The MIDA benchmark specifically tests multimodal LLMs' ability to detect deception in multi-person social interactions.
    • Experiments show that even advanced models with audio-visual capabilities struggle to accurately infer others' knowledge states and belief intentions.
    • The paper was accepted to the EMNLP 2026 Main Conference, with code and datasets open-sourced.
UTokyo Releases MIDA Benchmark Exposing Multimodal LLM Social Deception Blind Spots | Today for AI