Bessent Threatens Sanctions on Chinese AI Labs
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Treasury Secretary Scott Bessent threatened sanctions against Chinese AI labs on July 21, saying the government found "watermarks of our U.S. large language models on many of the Chinese models."
- White House science advisor Michael Kratsios clarified on July 22 that Moonshot AI built a purpose-built platform to extract Anthropic's Fable model while evading detection.
- Anthropic traced more than 3.4 million exchanges with its models to Moonshot by February 2026, routed through hundreds of fake accounts via an internal evasion platform.
- Anthropic told the Senate Banking Committee in June that Alibaba's Qwen lab conducted the "largest known distillation attack" against it to date.
- Nvidia's Jensen Huang said American companies should "absolutely" use Chinese models that outcompete on price, while investor Bill Gurley argued in the Washington Post against treating open models as a danger.
- Chinese open-source AI models now account for the majority of global AI workloads, complicating enforcement since open model weights cannot be unreleased.
Why it matters: The Trump administration is targeting fraud — not open-source AI — with Entity List and IEEPA sanctions aimed at cutting Chinese labs off from US cloud infrastructure and Western enterprise customers. Anthropic paid a $1.5 billion copyright settlement for training on pirated books in an American court; Moonshot and Qwen face no equivalent forum for extracting US frontier model weights, and sanctions aim to close that asymmetry.
