ValueFlow: Measuring the Propagation of Value Perturbations in Multi-Agent LLM Systems

2026-02-09

AI Summary

This work addresses the unclear mechanisms by which value perturbations propagate and induce value drift in multi-agent large language model (LLM) systems. The authors propose ValueFlow, a novel framework that decouples value perturbation propagation into agent-level response behaviors and system-level structural effects. They introduce two quantifiable metrics—β-susceptibility and System Sensitivity (SS)—to elucidate the micro- and macro-scale dynamics of value drift. Leveraging Schwartz’s theory of basic human values, they construct a 56-dimensional value dataset and employ an LLM-as-a-judge evaluation protocol within simulated multi-agent interactions. Extensive experiments across diverse model architectures, personality prompts, value dimensions, and network topologies demonstrate that different values exhibit markedly varying susceptibility to perturbations, and that system structure critically shapes both the pathways and intensity of value propagation.

Technology Category

• Humans and AI: Learning Human Values and Preferences
• Multiagent Systems: Adversarial Agents
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Multi-agent large language model (LLM) systems increasingly consist of agents that observe and respond to one another's outputs. While value alignment is typically evaluated for isolated models, how value perturbations propagate through agent interactions remains poorly understood. We present ValueFlow, a perturbation-based evaluation framework for measuring and analyzing value drift in multi-agent systems. ValueFlow introduces a 56-value evaluation dataset derived from the Schwartz Value Survey and quantifies agents'value orientations during interaction using an LLM-as-a-judge protocol. Building on this measurement layer, ValueFlow decomposes value drift into agent-level response behavior and system-level structural effects, operationalized by two metrics: beta-susceptibility, which measures an agent's sensitivity to perturbed peer signals, and system susceptibility (SS), which captures how node-level perturbations affect final system outputs. Experiments across multiple model backbones, prompt personas, value dimensions, and network structures show that susceptibility varies widely across values and is strongly shaped by structural topology.

Problem

• value propagation
• multi-agent LLM systems
• value drift
• value alignment
• agent interaction

Innovation

• ValueFlow
• value perturbation
• multi-agent LLM systems
• LLM-as-a-judge
• system susceptibility