[models] · · 1 min read
Treasury threatens sanctions over alleged AI model distillation by Chinese startup Moonshot
U.S. officials accuse Moonshot of distilling Anthropic's Fable model at scale, escalating the debate over IP theft and open-weight Chinese models.
By ByteBulletin Editors · Editorial Team
The U.S. Treasury is threatening sanctions against Chinese AI company Moonshot after White House officials accused the startup of conducting large-scale distillation of Anthropic's Fable model. Treasury Secretary Scott Bessent posted on X that "open source is not open season on American IP," warning that "sanctions and Entity List designations will be on the table" for firms that cross the line into IP theft.
The move follows an earlier statement from Bessent that the U.S. would examine Chinese open source models for signs of IP theft. White House science and technology policy chief Michael Kratsios specifically called out Moonshot, alleging the company acquired Nvidia GB300 servers (part of the Blackwell generation, banned for sale to Chinese firms) and accessed them in Thailand, likely to train its models. The GB300 servers are subject to U.S. export controls.
Model distillation is a common technique where a smaller model learns from a larger model's outputs. While often legitimate, it can also infringe on IP. Moonshot recently released its Kimi K3 model as open-weight, and its capabilities have raised questions about the business models of leading U.S. AI labs. Some experts doubt that K3 could have been developed primarily through distillation from Fable, which has only been publicly available since July 1.
This episode has reignited debate in Washington over the influx of Chinese open-weight models. Some, like former White House AI adviser Dean Ball (now at OpenAI), have argued for restricting or banning their use to preserve America's technological lead and mitigate national security risks.
SHARE
RELATED

[models] ·
Escha Labs ships 2-bit Qwen3.6-35B-A3B build that runs on a 24GB GPU
A 12.3GB MoE quant packs a 35B model onto consumer cards, with quality within noise of FP8 on most benchmarks.

[models] ·
ByteDance Is Training a 10 Trillion-Parameter Model, Leaning Into Scale to Chase Anthropic
The TikTok parent is reportedly pre-training a massive model that could rival Anthropic's Mythos 5, signaling a new phase in the global AI race.

[models] ·
Alibaba’s Qwen3.8-Max goes open-weight, claiming Claude-rivaling performance
Alibaba releases its largest open-weight model yet, Qwen3.8-Max, claiming it rivals Anthropic’s Claude Fable 5, with weights due next week.
