China Acts Out, America Orders the World to Respond

The United States Department of State has ordered its entire diplomatic network worldwide to take a stance regarding allegations of artificial intelligence (AI) technology theft by Chinese companies, including DeepSeek.

This instruction is contained in a diplomatic cable sent to US diplomatic and consular offices around the world. In the document, US diplomats are asked to convey concerns about China’s practices of extracting and distilling US-owned AI models.

“A separate request has also been sent to Beijing to be conveyed to the central government,” the document states, quoted from Reuters on Monday (27/4/2026).

Washington assesses that Chinese companies are training cheap AI models using outputs from more expensive and advanced US AI systems. This technique is known as distillation, which involves training small models from the results of large models to reduce development costs.

The White House has previously levelled similar accusations. Even OpenAI has warned US lawmakers that DeepSeek is attempting to replicate the American AI company’s models for use in training their own systems.

However, China strongly denies these allegations. The Chinese Embassy in Washington described claims of AI intellectual property theft as unfounded accusations and attacks on China’s AI industry development.

“Accusations that Chinese entities are stealing American AI intellectual property are baseless and constitute a deliberate attack on China’s development and progress in the AI industry,” the statement said.

Amid these accusations, DeepSeek has instead introduced a preview of its latest V4 AI model developed for Huawei chips. This move underscores China’s ambition to strengthen AI technological independence amid Western pressure.

Although several Western countries and parts of Asia have banned the use of DeepSeek due to data privacy concerns, the company’s model remains one of the most widely used on global open-source platforms.

In the cable, the US also warns that illegally distilled AI models may appear to have equivalent performance at much lower costs, but risk removing security protocols and system neutrality mechanisms.