The US State Department has issued a global directive to highlight what it describes as widespread attempts by Chinese companies, including AI startup DeepSeek, to pilfer intellectual property from U.S. artificial intelligence laboratories, as per a diplomatic cable obtained by Reuters.
The cable’s objective is to caution against the risks of utilizing AI models derived from U.S. proprietary AI models and to set the groundwork for potential future engagement by the US government.
Distillation refers to the process of training smaller AI models using outputs from larger, more costly ones to reduce the expenses of training a robust new AI tool.
DeepSeek, the Chinese startup that amazed the world with its cost-effective AI model last year, recently unveiled a preview of a highly anticipated new model tailored for Huawei chip technology, emphasizing China’s increasing independence in the field.
The State Department, DeepSeek, and the Chinese Embassy in Washington did not respond immediately to a request for comment. The cable also referenced Chinese AI companies Moonshot AI and MiniMax, both of which did not respond immediately to comment requests.
Recently, the White House leveled similar accusations, prompting the Chinese Embassy in Washington to dismiss them as “baseless allegations,” asserting that Beijing places great emphasis on safeguarding intellectual property rights.
The cable, dated Friday and distributed to diplomatic and consular missions worldwide, directs diplomatic personnel to engage with their foreign counterparts regarding concerns about adversaries extracting and distilling U.S. AI models.
“A separate demarche request and message has been dispatched to Beijing for discussion with China,” the document states.
The cable, previously unreported, indicates that the Trump administration is taking the mounting apprehensions about Chinese distillation of U.S. AI models seriously.
“A.I. models developed through covert, unauthorized distillation campaigns empower foreign entities to introduce products that seemingly perform comparably on specific benchmarks at a fraction of the cost but fail to replicate the full performance of the original system,” the cable mentions, noting that these campaigns also intentionally remove security protocols from the resultant models and dismantle mechanisms ensuring the AI models’ ideological neutrality and truth-seeking nature.
OpenAI cautioned U.S. lawmakers that Chinese AI startup DeepSeek was targeting ChatGPT maker and the country’s prominent AI firms to replicate models for its own training, as reported by Reuters in February.
The memorandum and subsequent cable, released shortly before U.S. President Donald Trump’s scheduled visit to Chinese President Xi Jinping in Beijing, are expected to escalate tensions in a protracted tech rivalry between the competing superpowers, which had been eased by a truce brokered last October.