Phantom Consensus: Social Reasoning Failures in Multi-Agent LLM Systems
arXiv — cs.AI · Disha Sheshanarayana, Rajat Subhra Pal, Manjira Sinha, Tirthankar Dasgupta · Published on Readzer
Summary
A study introduces PHANTOM-Bench, a benchmark measuring how ambiguous social signals degrade decision quality in multi-agent LLM collectives, leading to 'phantom consensus' where public decisions diverge from private judgments. Across nine models, public accuracy dropped by up to 47.8 percentage points, and collective rationality failures reached 47.3% in the most susceptible setting. Revealing agents' recorded private decisions repaired many failures without exposing the gold label, though recovery varied across models.
This is a reviewed summary. Read the complete article at the original source.
Why it matters by industry
- Artificial intelligence
Relevant to AI model evaluation and multi-agent systems, highlighting the need to measure private judgment alignment beyond final agreement.