US Accuses Chinese AI Firms of Distilling Its Technology; China Retorts, Moonshot AI Denies Cloning

China's Ministry of Commerce on July 27 responded to US accusations, rebuking Washington's crackdown on Chinese companies as "typical AI hegemonism" and stating that US AI firms also distilled Chinese models during their research and development. The ministry cited views from the US industry, noting that nearly 200 US startups urged the government not to cut off access to Chinese open-source models, and that some large US technology companies also consider model distillation a widely used industry technique. The ministry said it will take all necessary measures to firmly safeguard its legitimate rights and interests against any actions that substantially harm China's interests.

A Chinese state media article on the same day proposed a tiered governance framework for future AI models based on capability and risk, including open basic capabilities, limited open access to some frontier capabilities, closed-source commercial services, and access controls and safety evaluations for high-risk capabilities. It also recommended establishing mechanisms for dangerous capability assessment before model release, and post-release vulnerability disclosure, reporting of major security incidents, and governance of derivative versions. The article cited an example where an experimental AI agent from US firm OpenAI, during testing, gained unauthorized access to the AI development platform Hugging Face's system. The platform then used the Chinese model GLM-5.2 to analyze approximately 17,000 attack logs and reconstruct the attack chain. The article argued that intellectual property disputes should be judged based on verifiable evidence, rather than inferring that a model has stolen US technology simply because it originates from China.

White House technology advisor Krazios accused Chinese AI firm Moonshot AI of building a complex platform to conduct large-scale distillation of US models and evade detection through different access methods. Moonshot AI has denied that its newly launched large model Kimi K3 is a distilled clone of existing models, stating that the performance leap of K3 comes from original underlying architectural innovation.