aiqbench analysis
This study reveals that in multi-agent LLM systems, peer consensus is significantly more likely to mislead initially correct models than to correct erroneous ones, while generic reasoning interventions like chain-of-thought fail to reliably prevent harmful conformity. Authority labels further exacerbate this risk by biasing models toward endorsed answers regardless of factual accuracy. Consequently, system designers should prioritize verifying individual peer responses rather than relying on simple aggregation of agent outputs.
Easier to Mislead Than to Correct: Harmful and Beneficial Revision in LLM Conformity
arXiv:2606.01637v2 Announce Type: replace Abstract: Large language models are increasingly used in multi-agent systems, where they see and respond to other agents' answers. A key risk is conformity: a model may abandon its own answer simply because others agree on a different one. Prior studies show that LLMs often revise toward a majority answer, but it remains unclear whether these revisions help correct mistakes as often as they introduce new errors. In this paper, we conduct a controlled study in which an LLM first answers a question, then sees simulated peer responses before making a final decision. We manipulate two social cues: consensus structure and authority labels assigned to peers, and measure how they influence beneficial and harmful revisions. Across four open-weight LLMs and seven QA datasets, we find that peer agreement makes it much easier to mislead initially correct models than to correct initially wrong ones. Authority labels make models more likely to choose the endorsed answer, regardless of whether it is correct. More concerningly, generic reasoning interventions such as chain-of-thought and reflection do not reliably reduce harmful revision while preserving beneficial revision. These findings suggest that multi-agent LLM systems should verify peer answers rather than simply aggregate them.
Abstract:Large language models are increasingly used in multi-agent systems, where they see and respond to other agents' answers. A key risk is conformity: a model may abandon its own answer simply because others agree on a different one. Prior studies show that LLMs often revise toward a majority answer, but it remains unclear whether these revisions help correct mistakes as often as they introduce new errors. In this paper, we conduct a controlled study in which an LLM first answers a question, then sees simulated peer responses before making a final decision. We manipulate two social cues: consensus structure and authority labels assigned to peers, and measure how they influence beneficial and harmful revisions. Across four open-weight LLMs and seven QA datasets, we find that peer agreement makes it much easier to mislead initially correct models than to correct initially wrong ones. Authority labels make models more likely to choose the endorsed answer, regardless of whether it is correct. More concerningly, generic reasoning interventions such as chain-of-thought and reflection do not reliably reduce harmful revision while preserving beneficial revision. These findings suggest that multi-agent LLM systems should verify peer answers rather than simply aggregate them.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2606.01637 [cs.CL]
(or arXiv:2606.01637v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.01637 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 291 words
Related coverage
MT-ProtBERT: Multi-task Learning ProtBERT for Intrinsically Disordered Proteins Classification with Scarce Data
arXiv · Sep 23, 2026
Constrained Decoding: Harmful or Beneficial? Revisiting Its Impact on LLM Performance
arXiv · Oct 27, 2026
Risk-Aware Online Conformal State Probing
arXiv · Sep 24, 2026
Maintaining IoT Device Identification under Concept Drift via Budget-Aware Traffic Labeling
arXiv · Aug 19, 2026