US State Department orders global warning over Chinese AI ‘distillation’

TL;DR:

  • A previously unreported State Department cable tells US diplomatic posts worldwide to alert foreign counterparts to alleged Chinese “extraction and distillation” of US AI models, naming DeepSeek, Moonshot AI and MiniMax.
  • The cable hardens this week’s White House accusations into formal diplomatic messaging weeks before President Trump’s planned visit to Xi Jinping in Beijing.
  • For UK firms partnering with US AI labs, the practical signal is that export-control questions and downstream-model provenance are now an active diplomatic agenda item, not a theoretical one.

The US State Department has instructed diplomatic and consular posts worldwide to raise concerns over what it calls Chinese efforts to extract and distill US artificial intelligence models, according to a cable seen by Reuters and dated 24 April. A separate démarche has reportedly been routed to Beijing.

The cable formalises a line of accusation the White House began making publicly this week. It names DeepSeek explicitly and adds Chinese AI firms Moonshot AI and MiniMax. Distillation, in this context, refers to training smaller models using outputs from larger, more expensive ones — a technique that reduces the cost of building capable models without independently solving the underlying training problem. OpenAI told US lawmakers in February that DeepSeek had been targeting US AI labs to replicate models for its own training.

What the cable says

According to the text seen by Reuters, distillation campaigns “enable foreign actors to release products that appear to perform comparably on select benchmarks at a fraction of the cost but do not replicate the full performance of the original system.” It also alleges the campaigns “deliberately strip security protocols from the resulting models” and undo mechanisms intended to keep model behaviour neutral. The Chinese embassy in Washington called the accusations “groundless”.

DeepSeek, which has previously said its V3 model used naturally occurring web-crawled data and not synthetic OpenAI output, launched a preview of its V4 model the same day adapted to run on Huawei chips — underlining China’s parallel push for chip-stack autonomy.

Looking forward

The diplomatic timing is the part UK readers should pay attention to. The cable lands weeks before President Trump’s planned summit with Xi Jinping in Beijing, after a tech-war détente brokered last October. If the US escalates from cable-level messaging to formal export-control or sanctions action against Chinese AI firms or their downstream users, UK companies that have integrated open-weight Chinese models into production stacks will need to decide quickly whether their compliance posture survives a more aggressive US position.

Many Western and some Asian governments already prohibit institutional use of DeepSeek on data-privacy grounds. The next step — restrictions on commercial deployment by allies — is no longer an abstract risk.

Share this article