The US State Department has ordered a global diplomatic offensive to denounce what it describes as widespread maneuvers by Chinese companies, including the startup DeepSeek, to steal the intellectual property of American artificial intelligence laboratories, according to a diplomatic cable reviewed by Reuters.
The note specifies that its goal is to ‘warn of the risks associated with the use of AI models derived from distilling American proprietary models, and to pave the way for potential awareness-raising actions by the United States government.’
Distillation is a process of training lighter AI models using the results of larger and more expensive models, in order to reduce the development costs of a new high-performing tool.
DeepSeek, the Chinese startup whose low-cost model stunned the sector last year, unveiled a preview of a highly anticipated new model on Friday, tailored for Huawei chips, highlighting China’s increasing autonomy in this field.
The State Department, DeepSeek, and the Chinese embassy in Washington did not immediately respond to requests for comments.
The cable also mentions Chinese companies Moonshot AI and MiniMax. Neither of the two had an immediate reaction.
This week, the White House made similar accusations, which the Chinese embassy in Washington called ‘unfounded allegations,’ adding that Beijing ‘attaches great importance to protecting intellectual property rights.’
The document, dated Friday and transmitted to diplomatic and consular posts worldwide, instructs diplomatic personnel to address with their foreign counterparts ‘concerns regarding the extraction and distillation of American AI models by adversaries.’
‘A formal request and a separate message were sent to Beijing to be raised with China,’ the document states.
This cable, whose existence had not been revealed until now, indicates that the Trump administration is taking seriously the growing concerns about Chinese distillation of American AI models.
‘AI models developed from covert and unauthorized distillation campaigns enable foreign actors to market products that appear comparably performing on some benchmark tests for a fraction of the cost, but do not replicate the full capabilities of the original system,’ the note states, adding that these campaigns ‘deliberately suppress the security protocols of resulting models and distort the mechanisms ensuring that these AI models are ideologically neutral and truth-seeking.’
OpenAI had warned American legislators that the Chinese startup DeepSeek was targeting the creator of ChatGPT as well as the top AI leaders in the country to replicate their models and use them for its own training, Reuters reported in February.
This memo and the accompanying cable, released just weeks before the planned visit of President Donald Trump to Beijing to meet Xi Jinping, promise to exacerbate tensions in a longstanding technological war between the two rival superpowers, which had been eased by a negotiated détente last October.
