The AI Monoculture Paradox in Financial Markets

We study traders choosing between private human forecasts and a more accurate shared AI signal in a Kyle-type market. Full AI adoption can be the unique equilibrium while lowering traders’ combined expected profits: correlated errors and crowded trades can outweigh accuracy gains. In the continuum model’s paradox regime, an adoption tax with lump-sum rebates can implement the collective profit optimum. LLM-agent experiments show that beliefs about other traders can push adoption below equilibrium, while communication can produce full adoption even when theory predicts partial adoption, including attempts to mislead others.