US Accuses Chinese Firms, Including DeepSeek, of IP Theft as V4 Launches
The U.S. State Department ordered American diplomats to warn foreign governments about alleged large-scale IP theft by Chinese AI firms, hours before DeepSeek launched its V4 model on Huawei chips.
On April 24, 2026, the U.S. State Department issued a worldwide diplomatic warning, directing its offices around the world to speak with foreign governments about what Washington says are organized efforts by China to steal intellectual property from American AI labs.
The cable states that its purpose is to warn about the risks of using AI models distilled from U.S. proprietary AI models and to lay the groundwork for possible future actions and discussions by the U.S. government.
The document names DeepSeek explicitly, alongside Moonshot AI and MiniMax, with a separate diplomatic request sent directly to Beijing. Hours later, DeepSeek launched its most anticipated model, V4, making April 24 one of the most charged days in the U.S.-China AI rivalry.
What Distillation Is and Why Washington Calls It Theft
Model distillation, training smaller AI models on outputs from larger proprietary ones, sits at the center of the accusations.
As Mint confirmed, the White House alleged Chinese entities ran coordinated campaigns using tens of thousands of proxy accounts to evade detection systems at American AI companies.
They also used jailbreaking techniques to bypass safety controls and pull out proprietary model behavior, which could then be reused as training data.
White House OSTP Director Kratsios stated that Chinese entities are engaged in deliberate, industrial-scale campaigns to distil U.S. frontier AI systems.
The cable added that distilled models 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 frames distillation not as a technical shortcut, but as taking unfair advantage of American R&D investment.
OpenAI and Anthropic Had Already Filed the Same Accusations
Earlier, OpenAI told the U.S. House Select Committee on the Chinese Communist Party in February that these distillation attacks from Chinese groups continued even after it strengthened its defenses following DeepSeek V3.
In the same month, Anthropic accused three Chinese projects, including DeepSeek, of illicitly extracting data from its Claude system, describing the campaigns as growing “in intensity and sophistication.”
DeepSeek has previously denied intentional use of synthetic data from American models, stating its V3 training used data collected through organic web crawling. Neither DeepSeek, MiniMax, nor Moonshot AI responded to the State Department cable.
DeepSeek V4 Launches on Huawei Silicon the Same Day
Hours after the cable became public, DeepSeek released V4-Pro and V4-Flash, both built explicitly for Huawei’s Ascend AI chips, not Nvidia GPUs.
As Reuters confirmed, Huawei confirmed its full supercomputer system, Ascend SuperPoD line, now supports V4.
The hardware pivot is strategically pointed: DeepSeek’s earlier V3 relied on Nvidia processors, drawing accusations of export control violations. V4 eliminates that vulnerability.
This comes on the same day that Howard Lutnick confirmed that no advanced Nvidia chips have actually been shipped to China, despite earlier conditional approval in January, per Reuters.
On performance, V4-Pro carries 1.6 trillion parameters, trails only Google’s Gemini (Pro-3.1) among open-source models on world knowledge benchmarks. It also claims to offer the strongest agentic coding capability of any open-source release to date.
At $3.48 per million output tokens, against OpenAI’s $30 and Anthropic’s $25, the pricing gap remains one of the sharpest in the industry.
Beijing’s Response and What the Trump-Xi Summit Now Faces
China reportedly rejected every accusation. The Chinese Embassy in Washington called the allegations “groundless” and “a deliberate attack on China’s AI industry.” Beijing’s foreign ministry called the Kratsios memo “a smear” and urged Washington to “abandon biases.”
Analysts note the cable and Kratsios memo together mean AI intellectual property and chip supply controls are now live flashpoints heading into the scheduled Trump-Xi summit in Beijing.
The meeting was meant to strengthen a fragile agreement reached last October, but now it is expected to be more tense than both sides originally anticipated.
Source: White House memo claims mass AI theft by Chinese firms
