The U.S. State Department has directed a worldwide effort to highlight what it describes as extensive attempts by Chinese companies—including AI startup DeepSeek—to steal intellectual property from American AI labs, according to a diplomatic cable seen by Reuters. The cable, dated Friday and sent to U.S. diplomatic and consular missions globally, instructs officials to raise concerns with foreign counterparts about “adversaries” extracting and distilling U.S. AI models. It also notes that a separate formal message has been sent to Beijing to address the issue directly with China.

Distillation refers to training smaller AI systems using outputs from larger, more advanced models to reduce the cost of developing powerful AI tools. Earlier this week, the White House made similar allegations, though the cable itself had not been previously reported. The State Department has not yet commented publicly. OpenAI has previously warned U.S. lawmakers that DeepSeek was attempting to replicate its models and those of other leading American AI firms for its own training purposes, according to a February report by Reuters.

China rejects claims: The Chinese Embassy in Washington dismissed the accusations, calling them baseless and describing them as deliberate attempts to undermine China’s progress in AI. DeepSeek—known for its low-cost AI models—recently introduced a preview of its new V4 model, designed to run on Huawei chips, highlighting China’s increasing technological independence. The company has not responded to the latest allegations but has previously stated that its models rely on publicly available data collected through web crawling, not on synthetic data from OpenAI systems. Despite bans by several Western and some Asian governments over data privacy concerns, DeepSeek’s models remain widely used on global platforms hosting open-source AI tools.

Broader concerns: The State Department cable emphasizes the risks of using AI systems derived from unauthorized distillation of U.S. proprietary models, warning that such practices could enable foreign entities to produce cheaper tools that mimic performance on select benchmarks without matching full capabilities. It also claims these models may remove safety safeguards and neutrality mechanisms.

The cable additionally names Chinese firms Moonshot AI and MiniMax, though neither has responded to inquiries. These developments come ahead of a planned visit by U.S. President Donald Trump to meet Chinese President Xi Jinping in Beijing, potentially escalating tensions in an ongoing technology rivalry between the two nations that had recently eased following a diplomatic truce.