Updated
Updated · The Washington Post · Aug 4
US Alleges Chinese AI Labs Used 24,000 Fake Accounts to Distill American Models
Updated
Updated · The Washington Post · Aug 4

US Alleges Chinese AI Labs Used 24,000 Fake Accounts to Distill American Models

3 articles · Updated · The Washington Post · Aug 4

Summary

  • Moonshot AI, DeepSeek and MiniMax are accused by US officials and Anthropic of illicitly extracting American AI capabilities, with Moonshot’s Kimi K3 specifically alleged to have been built from Anthropic’s Fable.
  • Anthropic says the Chinese labs created about 24,000 fraudulent accounts and conducted 16 million exchanges with Claude, using fake accounts, masked routers and other methods to evade restrictions.
  • US officials argue the issue is not open-weight AI itself but provenance—distilling a company’s own model is standard practice, while secretly copying a rival’s finished system amounts to intellectual-property theft.
  • Washington casts the allegations as a strategic risk beyond AI pricing, warning that cheap Chinese models could become embedded across critical sectors and create dependencies similar to those it says emerged with Huawei and 5G.

Insights

Could the open-source AI model powering your business secretly be a stolen clone of a tech giant's proprietary system?
How can governments secure critical infrastructure when a public AI model can be poisoned with just ten malicious examples?