Increased LLM Capability Correlates with Higher Systemic Risk in Financial Markets

Frontier LLMs exhibit highly correlated reasoning, which creates non-diversifiable risk floors in agent-based simulations. While accurate reasoning reduces risk during high participation, shared misinformation environments trigger synchronized, correlated actions across capable agents.