According to a diplomatic cable reviewed by Reuters and dated Friday, U.S. embassies and consulates have been instructed to raise concerns with foreign counterparts regarding what Washington calls the “extraction and distillation of U.S. AI models by adversaries.” The directive forms part of a broader effort to shape international awareness and policy responses around emerging risks in advanced AI development.
At the center of the concerns is a technique known as distillation—a process in which smaller, less resource-intensive AI systems are trained using outputs generated by larger, more sophisticated models. While widely used in the industry to reduce training costs and improve efficiency, U.S. officials argue that the practice can also be exploited to replicate proprietary systems without authorization.
The cable also notes that a formal diplomatic message has been sent to Beijing to be raised directly with Chinese authorities.
The move follows similar allegations made earlier by the White House, which has accused Chinese AI developers of attempting to replicate American frontier models. The concerns have been echoed in private briefings to lawmakers, including warnings from OpenAI, which has reportedly flagged attempts by foreign actors to copy its systems for training competing models.
Among the companies mentioned in the diplomatic cable are Chinese AI firms DeepSeek, Moonshot AI, and MiniMax. U.S. officials allege that models developed through unauthorized distillation processes may appear competitive on benchmarks but fail to fully replicate the capabilities, safeguards, and alignment mechanisms of original systems. The document further claims such methods may strip away safety protections and weaken efforts to ensure neutrality and reliability in AI outputs.
In a swift response, the Chinese Embassy in Washington dismissed the accusations, stating that claims of intellectual property theft are “groundless” and politically motivated. It characterized the allegations as an attempt to undermine China’s progress in artificial intelligence and broader technological development.
DeepSeek, which gained global attention last year for its low-cost AI model, has recently continued to expand its offerings, including the preview launch of a new version reportedly adapted for Huawei chip infrastructure. The company has previously stated that its systems are trained using publicly available data and web-scraped information, and denies using proprietary outputs from competitors such as OpenAI.
The diplomatic escalation comes amid growing international scrutiny of DeepSeek and similar platforms, with several Western and Asian governments restricting or banning their use in official environments due to data security concerns. Despite this, the company’s models remain widely adopted on global open-source AI platforms.
The State Department cable emphasizes that its objective is to caution governments about reliance on AI systems derived from allegedly unauthorized model distillation, while also laying groundwork for further diplomatic engagement.
The issue adds fresh tension to an already strained technological rivalry between Washington and Beijing. It also arrives at a politically sensitive moment, as U.S. President Donald Trump prepares for a planned visit to Beijing to meet Chinese President Xi Jinping. Analysts suggest the renewed focus on AI intellectual property disputes could complicate efforts to stabilize relations between the two powers following a recent period of tentative diplomatic easing.