US Policy Shifts: Addressing Hostile AI Model Distillation by Foreign Actors (legrandcontinent.eu)

• White House memorandum formally accuses Chinese entities of industrial-scale AI model distillation.
• Anthropic and OpenAI evidence indicates millions of malicious queries to reverse-engineer model logic.
• U.S. businesses increasingly rely on cost-effective Chinese open-weight models, complicating containment strategies.

"The White House has issued a formal memorandum addressing 'Adversarial Distillation' of U.S. AI models, alleging that Chinese entities are systematically extracting proprietary capabilities from frontier models like Claude and GPT via massive proxy networks. With high-profile laboratories like Anthropic and OpenAI reporting millions of fraudulent interactions, the administration aims to facilitate intelligence sharing. However, the policy remains cautious regarding concrete sanctions, highlighting the tension between national security and the deep integration of Chinese open-weight models (like Qwen and Kimi) within the U.S. commercial stack, creating a complex dependency landscape."