US directs diplomats to flag Chinese AI

The U.S. State Department has told diplomats worldwide to warn foreign governments that Chinese artificial intelligence firms are allegedly siphoning know-how from American model makers. (reuters.com) A diplomatic cable dated April 24 and reviewed by Reuters instructed embassies and consulates to raise concerns about “extraction and distillation” of U.S. artificial intelligence models. Reuters said the cable specifically named DeepSeek and also mentioned Moonshot AI and MiniMax. (reuters.com) (cnbc.com) Distillation is a shortcut for building a new model by feeding it answers from a stronger one, like training a student by copying a teacher’s solved exam. U.S. officials say that method can reproduce capabilities without paying the original developer’s full research cost. (reuters.com) (openai.com) The cable marks a shift in Washington’s campaign on artificial intelligence. The U.S. had already tightened export controls on advanced chips, and now it is asking diplomats to carry its case directly to allied capitals. (reuters.com) This push follows a February memo in which OpenAI told U.S. lawmakers that DeepSeek was trying to replicate leading American systems and use those outputs for its own training. OpenAI said it had taken steps over the previous year to harden its models against distillation. (reuters.com) (openai.com) China rejected the accusations. Reuters reported that the Chinese Embassy in Washington called the claims “groundless” and said China opposes theft of trade secrets and protects intellectual property. (reuters.com) The timing is notable because DeepSeek on April 24 also unveiled a preview of its V4 model adapted for Huawei chips, underscoring China’s parallel push to build a domestic artificial intelligence stack. Reuters reported the model was tuned for Huawei’s Ascend processors rather than relying on Nvidia’s most advanced products. (reuters.com) Washington’s message to allies is no longer only about where Chinese firms get chips. It is also about how Chinese labs may be getting model behavior, and whether other governments will treat that as a trade, security, or enforcement issue. (reuters.com)