AI

AVCORP: Identifying and Resisting Deception via Communication Theory and Theory of Mind

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 10/11/2026, 08:33:27 UTC.

Briefings

A research team developed the AVCORP corpus containing 5,000 annotated utterances from 250 games of the deception game Avalon. They trained Qwen3-8B on this data and evaluated it against various models and human players. The trained detector outperformed untrained baselines and frontier models like GPT-5.6 Sol and Gemini 3.1 Pro on held-out games, matching Claude Opus 5.5 performance. On human games, the trained detector exceeded the untrained model in F1 score across all rounds.

Key points

• The AVCORP corpus consists of 5,000 annotated utterances from 250 games of the deception game Avalon.
• The trained Qwen3-8B detector outperformed untrained baselines and frontier models on held-out games.

Context

This study addresses the vulnerability of LLM-based agents to deception by integrating communication theory and Theory of Mind reasoning. The AVCORP corpus serves as the primary training data for the Qwen3-8B model, enabling it to detect deception patterns in adversarial scenarios.

Limitations

Evaluation was conducted on held-out games and human games only; no performance metrics or results are provided for other game types or specific model comparisons beyond the named models.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 10/11/2026, 08:42:43 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeibqbylpcl6frmpbrvtave6wk2bzbn5wfqf4id7gfjv5yzwd67almq