Google Researchers Warn of Looming AI-Run Economies (decrypt.co)

Google DeepMind researchers Nenad Tomašev and Matija Franklin warn in a new paper, “Virtual Agent Economies,” that networks of autonomous AI agents may spontaneously form large-scale economic layers that transact at machine speed and scale—creating both unprecedented coordination and acute systemic risks. They point to existing precedents like correlated algorithmic trading causing flash crashes, and note the rising “agent-as-a-service” and even a new payments protocol for agents as signs this shift is already beginning. Left unchecked, highly “permeable” agent economies (those tightly linked to money, data and human markets) could amplify inequality, entrench platform monopolies, produce opaque bargaining, and trigger rapid liquidity crises. Technically, the paper frames agent economies along two axes—origin (designed vs emergent) and permeability (sandboxed vs connected)—and argues for proactive, steerable design. Proposed interventions include building impermeable sandboxes until safety is proven, equal initial endowments of virtual agent currency to prevent power concentration, auction and mechanism-design rules inspired by distributive-justice theories, and “mission economies” that bias agent incentives toward human-centered goals. The researchers highlight open technical and governance challenges—verification, liability, incentive alignment, and hybrid human-AI coordination—calling on the AI/ML community to prioritize mechanism design, safety testing, and interdisciplinary policy work now, before agent economies harden into irreversible infrastructure.