U.S. warns of AI theft as Chinese models undercut American services

Treasury Secretary Scott Bessent revealed that American artificial-intelligence models carry deep-embedded identifiers, marking a pivot from the traditional hardware-centric U.S.-China competition to a software-focused confrontation. The market disruption caused by Chinese open-weight offerings such as Moonshot AI’s Kimi K3, which deliver advanced reasoning at minimal cost, has led enterprises to favor these inexpensive alternatives over domestic products.

Critics warn this mirrors the earlier Huawei/ZTE dumping strategy, where subsidized hardware flooded global networks before security risks emerged. The core of the threat lies in "adversarial distillation," where vast numbers of automated queries siphon the reasoning patterns of U.S. models like Claude and GPT, enabling foreign developers to train rival systems cheaply; Anthropic and OpenAI have confirmed such large-scale illicit querying.

In response, the House Committee on Foreign Affairs advanced the Deterring American AI Model Theft Act and the White House issued a memorandum designating the practice a national-security danger, paving the way for sanctions and export restrictions. Chinese authorities, meanwhile, are consulting domestic firms about possible export controls, suggesting the open-source narrative may be a temporary market-clearing tactic. The analysis underscores the asymmetry between China's state-driven, unrestrained approach and the United States' constitutional limits, which make rapid, blanket software bans disruptive to critical domestic users.