Token Taxes: mitigating AGI's economic risks

A new research paper titled 'Token Taxes: mitigating AGI's economic risks' proposes a novel fiscal policy to address the impending economic disruption of artificial general intelligence. Authored by Lucas Irwin, Tung-Yu Wu, and Fazl Barez, the paper argues that while AI safety research focuses on capability risks, the economic threats—including eroded tax bases, lowered living standards, and citizen disempowerment—are equally severe and understudied. The authors contend that the economic stagnation of the first industrial revolution would appear mild compared to a post-AGI world where human labor is largely obsolete, creating an urgent need for proactive policy frameworks.

The proposed 'token taxes' are usage-based surcharges on model inference, applied at the point of sale where AI is consumed, not where models are hosted. This design leverages existing compute governance infrastructure for enforcement through a staged audit pipeline involving black-box verification, norm-based tax rates, and white-box audits. The paper situates this approach within previous 'robot tax' debates but highlights two key advantages: enforceability and value capture at the usage point. For impact assessment, the authors call for agent-based modeling to simulate the economic effects and discuss alternative approaches like FLOP taxes and geopolitical challenges, including the risk of 'AI superpowers' vetoing such measures. The work was accepted at the prestigious ICLR 2026 Post-AGI Science and Society Workshop.