Alignment Drift in Persistent Multi-Agent LLM Simulations: A Theoretical Framework and Experimental Methodology

Mihai VanceaIndependent Research · April 2026Correspondence: vancea98 · GitHub: vancea98/gildedvoid Abstract We investigate whether large language model (LLM) agents, given persistent memory and iter