arXiv:2606.20493v1 Announce Type: new Abstract: When large language models serve as evaluators in multi-agent systems, their systematic evaluation biases propagate through the agent network. We introduce Contagion Networks, a formal framework for measuring how evaluator biases spread across interacting LLM agents. In a controlled 3-agent experiment using DeepSeek-chat with three distinct evaluator bias profiles (structured, balanced, evidence-based), we measure the Cross-Agent Contagion Matrix Gamma_3 and find that evaluator biases consistently propagate between agents (gamma in [0.157, 0.352]
The proliferation of multi-agent LLM systems for complex tasks makes understanding internal dynamics, particularly bias propagation, increasingly critical for reliable deployment.
This research provides a formal framework to quantify how biases in LLM evaluators can spread through agent networks, which directly impacts the trustworthiness and efficacy of AI-driven automation.
The understanding of LLM system vulnerabilities is deepened, requiring developers to account for recursive bias propagation in multi-agent architectures and potentially influencing future AI safety protocols.
- · AI Safety Researchers
- · Developers of Responsible AI Systems
- · Organizations prioritizing AI trustworthiness
- · Developers ignoring AI bias
- · Multi-agent system integrators without bias mitigation
- · LLM providers with uncharacterized evaluator biases
Increased focus on designing bias-resistant evaluation mechanisms and multi-agent interaction protocols.
Development of new tools and methodologies for detecting, measuring, and mitigating bias propagation within complex AI systems.
Regulatory bodies may begin to mandate formal bias propagation audits for multi-agent LLM systems in critical applications.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG