The Memory Curse: How Expanded Recall Erodes Cooperative Intent in LLM Agents

2026-05-08

AI Summary

This study investigates the impact of extended context windows and enhanced memory on cooperation in large language models (LLMs) within multi-agent social dilemmas. Contrary to intuition, the authors observe a systematic decline in cooperative behavior as memory capacity increases. Through repeated-game experiments across seven LLMs and four game types (28 configurations total), they identify a “curse of memory”: cooperation deteriorates primarily due to the *content* of stored interactions rather than memory length per se. Causal mechanisms are validated via reasoning trace analysis, synthetic “memory sanitization” (replacing harmful memories with benign ones), LoRA fine-tuning, and Chain-of-Thought ablation. Cooperation declines in 18 out of 28 settings, yet both prospective reasoning fine-tuning and memory sanitization significantly restore cooperative outcomes. Notably, these interventions exhibit zero-shot transferability across distinct games.

Abstract

Context window expansion is often treated as a straightforward capability upgrade for LLMs, but we find it systematically fails in multi-agent social dilemmas. Across 7 LLMs and 4 games over 500 rounds, expanding accessible history degrades cooperation in 18 of 28 model--game settings, a pattern we term the memory curse. We isolate the underlying mechanism through three analyses. First, lexical analysis of 378,000 reasoning traces associates this breakdown with eroding forward-looking intent rather than rising paranoia. We validate this using targeted fine-tuning as a cognitive probe: a LoRA adapter trained exclusively on forward-looking traces mitigates the decay and transfers zero-shot to distinct games. Second, memory sanitization holds prompt length fixed while replacing visible history with synthetic cooperative records, which restores cooperation substantially, proving the trigger is memory content, not length alone. Finally, ablating explicit Chain-of-Thought reasoning often reduces the collapse, showing that deliberation paradoxically amplifies the memory curse. Together, these results recast memory as an active determinant of multi-agent behavior: longer recall can either destabilize or support cooperation depending on the reasoning patterns it elicits.

Problem

• memory curse
• cooperative intent
• LLM agents
• social dilemmas
• context window expansion

Innovation

• memory curse
• multi-agent cooperation
• context window expansion
• forward-looking intent
• Chain-of-Thought reasoning