The U.S. State Department has ordered a global diplomatic push to warn foreign governments about what it calls widespread efforts by Chinese companies — including AI startup DeepSeek — to extract intellectual property from American artificial intelligence labs. The instruction was sent in a diplomatic cable dated Friday and obtained by Reuters, which broke the story on April 24, 2026.
The cable goes further than previous public warnings from U.S. agencies by naming specific Chinese AI firms and asking diplomats around the world to deliver the message in person to their counterparts.
What the cable instructs
According to Reuters, the cable was sent to diplomatic and consular posts globally, and tells staff to speak with foreign officials about "concerns over adversaries' extraction and distillation of US AI models." The document also names Moonshot AI and MiniMax alongside DeepSeek, and notes that "a separate demarche request and message has been sent to Beijing for raising with China."
Distillation, in this context, refers to training smaller, cheaper AI models using outputs sampled from larger, more capable ones. The technique is widely used in legitimate research, but U.S. officials argue that when applied to closed commercial models without authorization, it can effectively transfer hard-won capability from American labs to overseas competitors at a fraction of the original training cost.
Beijing pushes back
The Chinese Embassy in Washington dismissed the allegations in a statement to Reuters. "The allegations that Chinese entities are stealing American AI intellectual property are groundless and are deliberate attacks on China's development and progress in the AI industry," the embassy said.
The pushback is consistent with Beijing's broader position that U.S. export controls and rhetoric about model theft are tools of industrial competition rather than legitimate security policy. Chinese labs have repeatedly said their models are trained on permissible public data and their own engineering work.
Background: an escalating pattern
The diplomatic cable follows months of warnings from the U.S. private sector. Reuters reported in February that OpenAI had told U.S. lawmakers it believed DeepSeek was targeting it and other leading AI companies to replicate their models for use in DeepSeek's own training pipeline.
DeepSeek's V4 preview, released earlier this week, intensified the political backdrop by claiming open-source performance close to top U.S. systems while pricing inference at a fraction of frontier model rates — a combination that U.S. officials privately attribute, in part, to model distillation rather than purely independent research.
Why it matters
Moving the message from Washington memos to in-person diplomatic démarches is a significant escalation. It puts allied governments on notice that procuring or hosting models from named Chinese vendors could become a friction point in U.S. relations, much as Huawei did in the 5G era. For DeepSeek, Moonshot AI, and MiniMax, that could mean tougher conversations with European, Middle Eastern, and Southeast Asian customers who had begun integrating their models into local infrastructure.
It also raises the temperature ahead of any new round of U.S. export controls or sanctions on AI training compute — a lever the Trump administration has signaled it is willing to pull again if it concludes that distillation, rather than independent research, is closing the U.S.–China model gap.
