Document Type
Article
Publication Date
2026
Abstract
This essay introduces distillation laundering, a new form of regulatory arbitrage. Distillation, a technique for training large language models based on the outputs of other models instead of from scratch, is generally prohibited by the terms of service of closed-weight frontier models. However, it is economically efficient and has been key to the success of the cheap, distilled (largely Chinese) open-weight models, which have reshaped the global political economy of AI regulation. US companies have now begun to re-import the effects of distillation through various means—a process we call distillation laundering. We also discuss how attempts to lobby these models out of the market to protect closed business models face a barrier posed by their growing popularity among businesses and other users, which together form a powerful counter-interest group. Finally, we discuss potential equilibrium outcomes in AI governance and the rise of a Beijing Effect, where China’s ability to shape the deployment of open models is key to the global AI race.
Publication Citation
Forthcoming, Georgetown Journal of National Security Law & Policy (Volume XVII, 2026).
Scholarly Commons Citation
Edelson, Laura and Lancieri, Filippo, "Distillation Laundering: Regulatory Arbitrage and a Potential Beijing Effect in AI" (2026). Georgetown Law Faculty Publications and Other Works. 2720.
https://scholarship.law.georgetown.edu/facpub/2720
Included in
Business Organizations Law Commons, Intellectual Property Law Commons, International Law Commons, Law and Economics Commons, Law and Politics Commons, Legal Remedies Commons, National Security Law Commons