US Warns World on Chinese AI IP Theft Allegations

The US State Department has sent a diplomatic cable to posts worldwide instructing staff to raise concerns with foreign governments about alleged intellectual property theft by Chinese AI companies, including DeepSeek, Moonshot AI, and MiniMax. The cable, dated Friday and seen by Reuters, asks diplomats to warn counterparts about risks tied to AI models they say were built using unauthorised distillation of US proprietary systems. The White House made similar accusations earlier this week. Distillation means training a smaller, cheaper AI model using outputs from a larger, more expensive one. The State Department says this process, when done without permission, lets foreign firms release products that appear competitive on benchmarks but lack the full capability of the originals. The cable also claims such campaigns strip out safety and content-neutrality mechanisms from the resulting models. China's Embassy in Washington called the accusations groundless and a deliberate attack on China's AI development. DeepSeek has previously denied using synthetic data generated by OpenAI, saying its V3 model relied on web-crawled data. On Friday, DeepSeek released a preview of its new V4 model, built for Huawei chips, signalling continued progress independent of US hardware. The timing is sensitive: Trump is scheduled to visit Xi Jinping in Beijing within weeks, and the move risks reigniting tech-war tensions that had eased after an October détente. OpenAI has separately warned US lawmakers that DeepSeek was actively targeting leading American AI companies to replicate their models.