Memory Contagion: Cross-Temporal Propagation of Evaluator Bias via Agent Memory

2026-06-22

AI Summary

This work introduces and formally characterizes the phenomenon of “memory contagion,” revealing how evaluator-induced biases in agent memory propagate across time. Through controlled experiments across four phases, the study investigates two specific biases—length preference and authority bias—comparing idealized (oracle) and practical memory integration mechanisms. The findings demonstrate that even a low contamination rate of 0.2 leads to significant bias propagation. While memory integration mitigates length preference, it can inadvertently amplify authority bias, and no safe contamination threshold exists below which bias propagation is avoided. These results expose a critical vulnerability in current agent memory architectures, showing that biased inputs alone are sufficient to trigger cross-temporal bias transmission.

Abstract

Large Language Model (LLM) agents increasingly rely on memory systems to maintain long-term coherence. Recent work shows that agent memories degrade during continuous consolidation. However, existing research assumes memories are derived from unbiased experiences. In this work, we identify and formalize a novel phenomenon: Memory Contagion -- the cross-temporal propagation of evaluator bias through agent memory. We show that when agents are trained or guided by biased evaluators, their experiences become biased; when these trajectories are stored and consolidated into memory, the bias propagates to future agents retrieving from the same memory store, even when consolidation is perfect (oracle). Across two bias types (length preference, authority bias) and four experimental phases, we demonstrate: (1) Memory Contagion occurs even with perfect consolidation (oracle condition), proving that biased input is a sufficient cause of contagion; (2) Consolidation has opposite effects depending on bias type -- robustly attenuating length bias while preliminarily amplifying authority bias (single-run estimate), suggesting a bias-type-dependent interaction; (3) No observed safe threshold: bias propagation is detected at contamination rates as low as p=0.2. Our findings expose a critical vulnerability in current agent memory designs and provide formal tools for measuring cross-temporal bias propagation.

Problem

• Memory Contagion
• Evaluator Bias
• Agent Memory
• Bias Propagation
• Large Language Model

Innovation

• Memory Contagion
• Evaluator Bias
• Cross-Temporal Propagation
• LLM Agents
• Memory Consolidation