US Watermarks Found on Chinese AI Models, Raising IP Theft Concerns

Chinese AI systems caught with American watermarks

Chinese artificial intelligence models have been discovered containing watermarks from U.S. large language models, according to warnings from Treasury Secretary Scott Bessent and China analyst Gordon Chang during a Fox Business segment.

The allegations center on intellectual property theft, with specific evidence pointing to China's R1 model being distilled over ChatGPT-4. The discovery of American watermarks embedded in Chinese AI systems suggests direct copying or unauthorized use of U.S. technology rather than independent development.

Why it matters

This revelation underscores a critical vulnerability in the global AI race: even sophisticated watermarking techniques designed to protect proprietary models may not prevent determined state actors from reverse-engineering or copying American innovations. For U.S. technology companies investing billions in AI development, the theft represents both immediate competitive harm and a long-term national security concern as China leverages stolen intellectual property to accelerate its own AI capabilities.

Calls for economic consequences

Gordon Chang, speaking on the Mornings With Maria program, called on the Trump administration to impose harsher economic costs on Beijing in response to these violations. The discussion with host Maria Bartiromo emphasized the national security implications of China copying American technology, particularly in the strategically important artificial intelligence sector.

The watermark discovery provides concrete evidence of technology transfer that has long been suspected but difficult to prove. Large language model watermarks are typically embedded as technical signatures that identify the origin and ownership of AI systems.

Broader implications for US-China tech competition

The intellectual property theft allegations come as the United States and China compete for dominance in artificial intelligence development. American companies have invested heavily in creating advanced large language models, with significant resources devoted to both capability development and protection mechanisms.

The case highlights ongoing challenges in protecting digital intellectual property from state-sponsored theft, particularly when dealing with sophisticated actors capable of reverse-engineering complex systems. For policymakers, the incident raises questions about the adequacy of current export controls and technology protection frameworks.

These details were first reported by Fox Business during their August 6, 2026 broadcast.