U.S. Treasury Secretary Scott Bessent warned that sanctions are possible after White House officials accused Chinese AI company Moonshot of improperly distilling Anthropic's Fable model to develop its own K3 model. The White House alleged Moonshot obtained Nvidia GB300 servers banned from export to China and conducted large-scale distillation against U.S. AI models. The accusations have intensified debate in Washington about whether the U.S. should restrict Chinese open-weight AI models to preserve technological advantage.
White House official Michael Kratsios accused Chinese AI startup Moonshot of using deceptive methods to extract data from Anthropic's Fable 5 model and use it to train its Kimi K3 system, framing model distillation as technology theft rather than competitive practice. Kratsios alleged Moonshot used a purpose-built platform with multiple access methods to avoid detection and acquired Nvidia GB300 servers in Thailand to train K3, a 2.8 trillion-parameter model released six days after the accusations. The administration's claims could trigger tighter API restrictions, expanded export controls on AI chips, and sanctions on Moonshot, though the allegations lack publicly verifiable evidence and the companies have not commented.
Nathan Lambert and Florian Brand discuss the rapid advancement of Chinese AI models like Kimi K3 and Qwen, analyzing performance gaps between open and closed models and the role of post-training optimization. They cover benchmarking challenges, infrastructure requirements for deploying large models, and how the open-model ecosystem is maturing in infrastructure and tooling. The conversation explores how open-weight models can be fine-tuned for specific tasks and the geopolitical implications of China's commitment to open-source AI development.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
