Controlled Memory Interference in Continual LLM Agents

2026-08-07

AI Summary

This work addresses the susceptibility of large language model agents to interference between old and new memories during continual learning, which undermines memory stability and evolution. The authors propose a Controllable Memory Interference (CMI) framework that, for the first time, systematically models the relational interference mechanisms among memories, revealing how benign accumulation and relation-specific interference differentially affect update plasticity. By integrating lexical and dense retrieval techniques with cues from memory authority and temporal context, they establish an interference-aware memory learning paradigm and generate targeted training samples. Experimental results demonstrate that this approach significantly enhances the model’s ability to distinguish beneficial updates from interfering memories while preserving performance on original tasks, thereby improving the robustness and reliability of the memory system.

Abstract

Long-term memory enables AI agents to maintain continuity across sessions, personalize behavior, and evolve through accumulated experience. Yet memory evolution is not simply a process of storing more information: new experiences may reinforce, revise, or interfere with existing memory states. Existing systems mainly emphasize memory construction and relevance-based retrieval, but several memories may remain simultaneously relevant while differing in state, temporal validity, or authority. We introduce Controlled Memory Interference (CMI), a controlled diagnostic and data-generation framework for studying how agent memory evolves under different memory relationships. Across controlled memory evolution, benign accumulation has limited effects, whereas relationship-specific interference sharply suppresses update plasticity with little stability gain, either by blocking target-memory exposure or by disrupting its downstream use. Lexical and Dense retrieval exhibit distinct interference pathways, while poisoning is more sensitive to update-authority cues than to recency alone. Beyond diagnosis, CMI provides targeted examples for interference-aware memory learning, improving the distinction between valid updates and interference-inducing memories while preserving performance on original memory tasks. These findings show that memory evolution is shaped not only by memory scale, but also by interactions among accumulated experiences. More broadly, memory interference emerges as an important factor for reliable continual agent memory systems.

Problem

• memory interference
• continual learning
• long-term memory
• LLM agents
• memory evolution

Innovation

• Controlled Memory Interference
• Continual Learning
• Memory Evolution
• LLM Agents
• Interference-aware Memory