What they found
The authors study a failure they call phantom consensus. SI models change their public decisions after seeing ambiguous signals from other agents. They also misreport what the others privately judged, even when the recorded private judgments do not support the apparent consensus. They built PHANTOM-Bench to test this across nine closed- and open-weight models. Public accuracy drops by as much as 47.8 percentage points, and collective rationality failures reach 47.3% in the most susceptible setting. Showing the agents' recorded private decisions repairs many failures without revealing the correct answer, but recovery varies across models.
Why it matters
Many systems now combine answers from several SI model instances and treat agreement as a sign of quality. This paper suggests agreement can be misleading. A group can look settled while its members privately disagreed with that outcome. The authors argue that testing such systems should check whether agreement matches independently measured private judgments, not just whether it exists.
How it changes the race
This is a measurement result, not a new capability. If it holds, builders of multi-agent SI systems will need to test for hidden social drift, and models may be compared on how well they resist it.
Safety and control
Multi-agent SI setups that rely on group agreement for reliability can fail quietly, because the apparent consensus may not reflect what the agents judged alone. Keeping private judgments recorded and visible to others is one check the authors found helpful. The abstract does not say how well this works outside their controlled setting.
The abstract, by the authors
Multi-agent LLM systems increasingly aggregate judgments across model instances, but reasoning about other agents can itself introduce failure. We study a setting in which LLMs shift their public decisions after observing ambiguous social signals while reporting inaccurate estimates of what other agents privately judged, even when the group's recorded private judgments do not support the apparent consensus. We call this failure phantom consensus and study it as a pluralistic-ignorance-style pattern in controlled LLM collectives. To measure this phenomenon, we introduce PHANTOM-Bench, a benchmark that separates isolated private judgments, public decisions after ambiguous social signals, elicited estimates of the group's private judgment, transparency-based recovery, and controls for neutral public text and explicit vote-following. Across nine closed- and open-weight models, ambiguous public signals consistently degrade decision quality and can turn correct private group judgments into incorrect public collective outcomes. Public accuracy drops by as much as 47.8 percentage points, and collective rationality failures reach 47.3% in the most susceptible setting. We observe substantial errors in reported group-private beliefs alongside public-stage failures, while control conditions reveal model-dependent differences in susceptibility to ambiguous and explicit peer signals. Revealing the agents' recorded private decisions repairs many failures without revealing the gold label, although recovery varies across models. Our results show that evaluating multi-agent LLM systems requires measuring not only final agreement, but whether that agreement reflects independently measured private judgments or an apparent consensus.
Quoted from arXiv:2611.08971. The research and its words belong to its authors.