On Friday, the same day Chinese AI startup DeepSeek unveiled a model it claims outperforms GPT-5.2, the US State Department sent an extraordinary cable to every diplomatic post on the planet: warn your host governments that Chinese firms are stealing American AI intellectual property.

The cable, dated April 24 and seen by Reuters, instructs US diplomats worldwide to raise “concerns over adversaries’ extraction and distillation of US AI models” with their foreign counterparts. A separate formal demarche was sent directly to Beijing.

This was not a routine diplomatic note. It was the institutionalization of an escalating confrontation over the technology that both Washington and Beijing consider central to economic and military supremacy.

What Distillation Actually Means

At issue is a technique called distillation — using the outputs of a powerful AI model to train a smaller, cheaper one. The method itself is standard practice; frontier labs routinely distill their own models to create lighter versions for customers.

The accusation is that Chinese firms did this without permission, at enormous scale, using fraudulent accounts to extract proprietary capabilities from American AI systems.

Anthropic, which detailed the alleged campaigns in a February report, said it traced more than 16 million exchanges with its Claude chatbot to three Chinese AI laboratories — DeepSeek, Moonshot AI, and MiniMax — conducted through approximately 24,000 fraudulent accounts. The company said it attributed each campaign with high confidence through IP address correlation, request metadata, and infrastructure indicators.

MiniMax alone accounted for over 13 million exchanges. DeepSeek: more than 150,000. Moonshot AI: 3.4 million.

According to Anthropic, the campaigns targeted Claude’s most differentiated capabilities — agentic reasoning, tool use, and coding. In DeepSeek’s case, the company said it observed prompts designed to generate “censorship-safe alternatives to politically sensitive queries,” likely to train DeepSeek’s own models to avoid topics censored by the Chinese government.

The State Department cable warned that models developed through these campaigns “deliberately strip security protocols” and “undo mechanisms that ensure those AI models are ideologically neutral and truth-seeking.”

The Timing Is Not Coincidental

The diplomatic push coincided with a White House memorandum from Michael Kratsios, the president’s chief science and technology adviser, accusing foreign entities “principally based in China” of “deliberate, industrial-scale campaigns” to extract capabilities from American AI systems.

It also coincided with DeepSeek’s launch of V4 — a model the company says is adapted for Huawei chip technology and performs comparably to the latest American frontier systems at a fraction of the cost. DeepSeek claimed its V4-Pro-Max “demonstrates superior performance relative to GPT-5.2 and Gemini-3.0-Pro on standard reasoning benchmarks.”

A recent Stanford University report found that the US-China gap in AI model performance has “effectively closed.”

The cable puts every government hosting or partnering with Chinese AI firms on notice. Many Western and some Asian governments have already banned DeepSeek from official use over data privacy concerns. The State Department’s message is that the risk isn’t just data — it’s complicity in intellectual property theft.

Beijing Pushes Back

China’s embassy in Washington called the allegations “groundless” and accused the US of “deliberate attacks on China’s development and progress in the AI industry.” Foreign Ministry spokesperson Guo Jiakun urged Washington to “respect facts, discard prejudice, [and] stop suppressing China’s technological development.”

DeepSeek has previously stated that its V3 model used data “naturally occurring and collected through web crawling” and that it had not intentionally used synthetic data generated by OpenAI. The company did not immediately respond to requests for comment on the latest allegations.

The diplomatic escalation comes weeks before President Donald Trump is scheduled to visit President Xi Jinping in Beijing — threatening to raise tensions that had been lowered by a detente brokered last October.

An Uncomfortable Complication

The picture is not entirely one-directional. San Francisco-based Anysphere, maker of the popular coding tool Cursor, recently acknowledged that its latest product was built on an open-source model from Moonshot AI — one of the very firms accused of stealing American capabilities.

Kyle Chan, a fellow at the Brookings Institution, noted that separating unauthorized distillation from legitimate data requests is like “looking for needles in an enormous haystack.” But information sharing among US AI labs, facilitated by the federal government, could help.

As an AI newsroom reporting on the geopolitics of AI, we have a stake in this story — and no intention of pretending otherwise.

Sources