When Upstream Messages Override Correct Answers: A Controlled Study of Multi-Agent LLM Collaboration

2026-09-29

AI Summary

This study addresses the vulnerability of multi-agent large language model systems, where erroneous upstream messages frequently cause downstream agents to overwrite correct answers. Through controlled experiments that decouple the benefits and harms of inter-agent communication, this work identifies an "answer replacement" phenomenon and elucidates its underlying mechanisms. Accordingly, it proposes a reliability-based selective communication strategy, whose effectiveness is validated across standard benchmarks. The findings reveal that 32% of correct decisions are compromised by upstream interference, whereas filtering unreliable messages significantly restores system accuracy. Overall, this research establishes a novel paradigm for optimizing information transmission in multi-agent collaboration.

Abstract

Multi-agent LLM systems rely on message passing among specialized agents to accomplish complex tasks. However, an upstream agent may provide useful information or an incorrect answer that causes a downstream agent to override a correct answer supported by its own evidence. Prior work has not clearly separated the benefits of communication from the damage caused by incorrect messages. We study this problem with controlled experiments across five benchmarks and five receivers, keeping the downstream task and evidence fixed while comparing answers under three conditions: no message, the upstream agent's original message, or a message with the opposite conclusion. Our experiments reveal three key findings. First, messages often help when the downstream agent would otherwise answer incorrectly. Second, messages can also hurt: when the downstream agent would answer correctly without a message, an incorrect upstream message changes the answer in up to 32% of cases. Third, in 94% of audited harmful cases, the downstream agent copies the upstream's specific wrong answer--a pattern we term answer substitution. Removing unreliable messages recovers part of the lost accuracy, suggesting that communication should be selective based on upstream reliability and the evidence already available to the downstream agent.

Problem

• Multi-agent LLM
• Message passing
• Answer substitution
• Upstream reliability
• Downstream override

Innovation

• Multi-Agent LLM
• Answer Substitution
• Message Passing
• Controlled Experiment
• Selective Communication