← Back

Wire · regulatory

China's Moonshot AI stole from Anthropic, Trump tech adviser says

Published

23 July 2026

Topic

regulatory

Sectors

AI Frontier Models

Geography

ChinaUnited States

Source

Read at bbc.co.uk

Verified

Fusion42 · 23 July 2026 · Fusion42 review

US officials have accused China's Moonshot AI of systematically stealing capabilities from Anthropic's models through distillation attacks and illicit access to restricted Nvidia servers to develop its K3 model. The Treasury Secretary has signalled sanctions are being considered for Chinese companies conducting industrial-scale IP theft of American AI technology.

This Wire brief sits within Fusion42's coverage of AI Frontier Models, and 14 sources have reported it between 22 Jul 2026 and 6 Aug 2026.

◆ The Wire takeaway

If you're selling AI models to US government or enterprise, you now need contractual controls on distillation and IP extraction - the White House is moving from warnings to sanctions enforcement, starting with Chinese competitors but the regime applies to anyone. Moonshot's open-source K3 release on 27 July becomes a test case for whether your model architecture can be legally protected from this attack.

Coverage

14 sources · first reported 22 Jul 2026 · latest 6 Aug 2026

Topics

AI Frontier Modelsai-securityip-theftchina-us-techsanctions-threatdistillation-attacks