Feed
researchAI Security and Safety11d ago

Researchers Introduce AVCORP Corpus to Enhance AI Agent Deception Detection via Theory of Mind

Facts

Researchers have released AVCORP, a dataset of 5,000 annotated utterances from the game Avalon, designed to train AI agents in deception detection using Theory of Mind and communication theory. By fine-tuning Qwen3-8B on this data, the researchers achieved performance in identifying deceptive tactics that matches state-of-the-art frontier models.

  • ▸AVCORP consists of 5,000 annotated utterances mapped to 37 specific deception tactics.
  • ▸The dataset includes reasoning traces covering abductive hypotheses, suspicion levels, and nested Theory of Mind beliefs.
  • ▸Qwen3-8B fine-tuned on AVCORP outperformed untrained baselines and matched Claude Opus 5.5 in deception detection benchmarks.
  • ▸The model showed improved F1 scores against both AI and human gameplay scenarios.
  • ▸The corpus aims to improve agent robustness against adversarial communication.
Occurred
2026-11-06 19:57 UTC
First reported
2026-11-06 19:57 UTC
Last updated
2026-11-10 03:25 UTC
Event type
research
Region
Global

Entities

Categories

Source timeline · 1

  • arXivFirst-party2026-11-06 19:57
Community vote

Sign in to vote on importance and sentiment.

Platform Signals

Impact score
68
AI sentiment
Positive
Event significance30%1
Entity importance20%1
Source credibility20%1
Novelty15%1
Corroboration15%1
Confidence1
Novelty1
Corroboration1
Community
ImportanceNo votes
Strategic sentiment
Positive 0%Neutral 0%Negative 0%

Platform signals are algorithmic and community-derived indicators. They are not investment recommendations and do not predict valuation or performance.