U.S. warns allies about Chinese firms accused of stealing AI models

The State Department has ordered diplomats worldwide to tell foreign counterparts that China’s AI push now includes large-scale efforts to extract value from American models and intellectual property, with DeepSeek, Moonshot AI and MiniMax singled out in the warning. The cable turns a software fight into a formal foreign-policy campaign, casting “risks of utilizing AI models distilled from U.S. proprietary AI models” as a national-security problem, not just a dispute between companies.

The instruction, sent to U.S. diplomatic and consular posts on April 24, followed a White House memo one day earlier from Michael Kratsios, the director of the Office of Science and Technology Policy. Kratsios said foreign entities, principally based in China, were running “industrial-scale” campaigns to distill U.S. frontier AI systems. The administration said it would share information with American AI companies and explore ways to hold foreign actors accountable, a sign that Washington is trying to coordinate pressure across diplomacy, industry and enforcement.

The evidence chain Washington is putting in front of allies rests heavily on OpenAI’s own warnings to lawmakers. In a February 12 memo to the House Select Committee on Strategic Competition between the United States and the Chinese Communist Party, OpenAI said DeepSeek’s next model should be understood in the context of “ongoing efforts to free-ride” on OpenAI and other U.S. frontier labs. OpenAI said it had seen attempts to distill frontier models using new, obfuscated methods, and had separately warned that DeepSeek employees were trying to bypass access restrictions and obtain model outputs programmatically for distillation. OpenAI also tied China’s AI drive to Beijing’s stated goal of becoming the global leader in AI by 2030.

That broader contest is no longer limited to chips and cloud capacity. DeepSeek has become a symbol of China’s effort to narrow the gap with U.S. AI firms, and Reuters reported this week that the company previewed a new model adapted for Huawei chip technology. The message from Washington is that the next stage of decoupling is not only about where semiconductors are made, but about who controls models, training data and the safety guardrails built into them.

The diplomatic push comes after the State Department announced its first sanctions under the Protecting American Intellectual Property Act on February 24, targeting one individual and two entities and saying trade-secret theft costs U.S. industry billions of dollars each year. That action was linked to a separate case involving Australian national Peter Williams, who pleaded guilty in October 2025 to theft of trade secrets. China’s Embassy in Washington rejected the allegations as baseless, and Beijing’s foreign ministry urged the United States to abandon bias and promote scientific and technological exchanges instead. With a summit between Donald Trump and Xi Jinping ahead, the AI dispute is becoming another test of how far Washington is willing to turn techno-diplomacy into leverage.