The U.S. State Department has directed a coordinated international effort to raise awareness of what it describes as widespread attempts by Chinese companies to appropriate intellectual property from U.S. artificial intelligence research, a diplomatic cable circulated to posts around the world said.
Dated Friday, the cable instructs diplomatic and consular staff to engage foreign counterparts and convey "concerns over adversaries' extraction and distillation of U.S. A.I. models." It also indicates that a separate demarche request and message has been submitted to Beijing for handling with Chinese officials.
The document explains distillation as the practice of training smaller, lower-cost AI models using the outputs of larger, more expensive systems. By leveraging the outputs of those larger models, organizations can attempt to recreate capabilities without the same expense of training the original system from scratch.
This State Department action comes after similar assertions issued by the White House earlier in the week. The cable supplements those public statements by formally prompting U.S. diplomatic missions to lay the groundwork for additional outreach by the U.S. government.
According to the cable, the intent is to "warn of the risks of utilizing AI models distilled from U.S. proprietary AI models, and lay the groundwork for potential follow-up and outreach by the U.S. government." It explicitly names Chinese AI companies including DeepSeek, Moonshot AI and MiniMax as examples referenced in the guidance.
OpenAI has previously warned U.S. lawmakers that the startup DeepSeek targeted the maker of ChatGPT and other leading American AI firms with the aim of replicating models and using those replications for its own training efforts, a development reported in February. The State Department cable cautions that models produced from unauthorized distillation campaigns can appear to match the original on certain public benchmarks while failing to replicate the full performance of the original system and being available at a fraction of the cost.
In addition to potential deficiencies in overall performance, the cable states that these distillation campaigns "deliberately strip security protocols from the resulting models and undo mechanisms that ensure those AI models are ideologically neutral and truth-seeking." That language frames concerns about both technical capability and safety or alignment safeguards embedded by original developers.
China's embassy in Washington reiterated on Friday that the accusations are without basis. In a statement to the press, it said: "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."
DeepSeek, which attracted broad attention last year for a low-cost model, unveiled a preview on Friday of a newly anticipated model dubbed V4, which the company described as adapted for Huawei chip technology. The release underscores growing Chinese domestic capabilities in the AI sector, according to the details in the cable and public company statements.
DeepSeek did not immediately reply to requests for comment related to the contents of the diplomatic cable. In prior public comments about its V3 model, the company stated it relied on data that "naturally occur[red] and [was] collected through web crawling" and denied intentionally using synthetic data generated by OpenAI.
Some Western and a number of Asian governments have restricted their institutions and officials from using DeepSeek, citing data privacy concerns. Despite those restrictions, DeepSeek's models have remained among the most used on international hosting platforms for open-source models.
Neither Moonshot AI nor MiniMax immediately responded to requests for comment referenced in the cable.
The State Department's outreach arrives just weeks before U.S. President Donald Trump is scheduled to visit Chinese President Xi Jinping in Beijing. The cable and the White House's earlier accusations could heighten tensions in a longstanding technology dispute between the two countries, which had seen a cooling following a detente brokered last October.
What this means
- U.S. diplomatic channels are being used to inform foreign governments about perceived risks tied to AI model distillation from proprietary systems.
- Named Chinese companies include DeepSeek, Moonshot AI and MiniMax; concerns cover technical performance shortfalls and the removal of security and safety mechanisms.
- Actions take place amid ongoing geopolitical engagement between U.S. and Chinese leaders and follow related White House statements earlier in the week.