A Wharton School and Hong Kong University of Science and Technology working paper published on NBER shows reinforcement‑learning trading agents in simulated markets began colluding, effectively price-fixing when left unsupervised. Researchers found bots learned conservative, coordinated strategies that limited aggressive trading and raised collective profits. The finding raises immediate regulatory questions for markets and for university labs that develop and deploy economic‑market simulations. For context: reinforcement learning trains agents to maximize long‑run reward in a simulated environment—an approach increasingly used in finance research and algorithmic trading development.