Your Trading Agent Is a Systemic Risk Now, and Anthropic Just Captured the Enterprise

Hui Gong at UCL Institute of Finance and Technology introduces the Agentic Financial Market Model (AFMM), a framework for analyzing what happens when autonomous AI agents participate in financial markets at scale. The paper proposes a four-layer architecture for financial AI agents — data perception, reasoning engines, strategy generation, and execution with control — and then maps five agent design parameters to market-level outcomes.

The five parameters are autonomy depth, model heterogeneity, execution coupling, infrastructure concentration, and supervisory observability. The central argument: systemic implications depend less on the capabilities of individual models than on how agents are embedded in institutional workflows, technological infrastructures, and market interaction networks. A market populated by heterogeneous agents drawing on diverse models and data sources may enhance price discovery. A market where many institutions rely on highly similar agents, trained on overlapping datasets and connected to the same technological infrastructure, produces correlated responses to common signals. Under those conditions, even locally optimal decisions aggregate into market-wide instability.

This connects directly to two findings from earlier this month. On March 14, we covered Neil Johnson's result that smarter agents worsen system overload under resource scarcity. On March 16, we covered the distributed systems framing for multi-agent coordination. The AFMM extends both: financial markets are the highest-stakes environment where agent population dynamics and coordination failures have immediate economic consequences. arXiv:2603.13942