AI RESEARCH
The Memory Curse: How Expanded Recall Erodes Cooperative Intent in LLM Agents
arXiv CS.AI
•
ArXi:2605.08060v1 Announce Type: cross 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.