State Department Names Three Chinese AI Firms in Escalating IP Theft Dispute

The U.S. State Department has taken an unusually direct step in the ongoing AI technology rivalry, issuing a diplomatic cable that explicitly names Chinese AI companies DeepSeek, Moonshot AI, and MiniMax in connection with systematic theft of American artificial intelligence models. The April 24 directive instructs U.S. diplomats to warn foreign governments about what it calls “adversaries’ extraction and distillation of U.S. AI models.” This marks a shift from general White House warnings to specific corporate accusations.

Distillation is not a lab curiosity. It is a method where competitors bombard U.S. AI systems with millions of queries through fake accounts and proxies, then train cheaper imitations on the responses. The costs drop dramatically, benchmarks often match, but safety guardrails frequently disappear. Anthropic detailed the practice in February, reporting that DeepSeek, Moonshot, and MiniMax operated 24,000 fraudulent accounts that generated over 16 million exchanges with Claude. MiniMax alone accounted for 13 million. Moonshot targeted agentic reasoning, coding, and vision tasks with 3.4 million queries.

OpenAI has echoed the alarm, telling lawmakers that DeepSeek is “free-riding” on its technology. White House Office of Science and Technology Policy Director Michael Kratsios sent a memo to federal agencies stating that “foreign entities, principally based in China, are engaged in deliberate, industrial-scale campaigns to distill U.S. frontier AI systems.” Proxy accounts evade bans; jailbreaks expose model internals. The result: Chinese models like DeepSeek-V3 and R1 now rival ChatGPT at a fraction of the cost, sidestepping U.S. chip export controls.

China’s embassy in Washington pushed back, stating that Beijing “attaches great importance to the protection of intellectual property rights.” None of the named companies commented to Reuters. The timing is pointed—President Trump is scheduled to meet with China’s Xi Jinping in May, and tensions are rising ahead of that summit.

The technique works. Train a small model on a large model’s outputs, and you bypass the need for massive datasets or expensive GPU clusters. U.S. firms invest billions in frontier systems; Chinese competitors skip that bill and launch market disruptors. DeepSeek’s V4 preview now runs on Huawei Ascend chips, signaling self-reliance. Yet insiders note these copies struggle with complex reasoning, hallucinate more frequently, and lack generalization.

The State Department cable sets the stage for coordinated action: “Warn of the risks of utilizing AI models distilled from U.S. proprietary AI models, and lay the groundwork for potential follow-up.” A formal demarche has already been delivered to Beijing. Lawmakers are examining violations under the Economic Espionage Act and the Computer Fraud and Abuse Act. Bipartisan sanctions bills are advancing, and entity list additions for the accused firms appear imminent.

Google flagged over 100,000 prompts cloning Gemini earlier this year—mostly from China. OpenAI saw most of its extraction attacks originate there as well. Anthropic blocked the flood but warned that distillation strips safety features, increasing risks for misuse in cybersecurity and bioweapons development. The White House characterizes the practice as a direct bypass of export controls.

Industry leaders are feeling the pressure. U.S. labs profit from API fees during these attacks—tens of millions of dollars—but lose their competitive edge. Chinese models flood global markets at low prices. DeepSeek tops app stores. Moonshot’s Kimi and MiniMax’s offerings are gaining traction. Yet without original training data, their limits show: benchmarks shine, but real-world depth remains shallow.

So what comes next? Diplomats are fanning out. Allies will be told to shun distilled products. U.S. companies are coordinating defenses—rate limits, anomaly detection, geo-fencing adjustments. Labs like Anthropic and OpenAI have already hardened their systems. But the scale is staggering: 16 million queries in a single campaign. Industrial.

The Trump administration is shifting focus from chips to models. The OSTP memo commits to protecting innovation, sharing tactics with firms, and exploring legal avenues. Ahead of the May summit, this hardens negotiating lines. China denies. The U.S. doubles down. The AI cold war is heating up, and models are now treated as munitions.

Fragmented defenses won’t suffice. Global posts are amplifying the warning. Risks include cheap copies without controls, national security exposure, and enterprises unknowingly adopting tainted models. Benchmarks deceive. True capabilities require scrutiny.

DeepSeek launched V4 on Friday amid the storm—aggressive pricing, Huawei compatibility. Defiance or coincidence? The State Department has named them the lead actor. Moonshot and MiniMax follow. The cable is clear: this theft undercuts U.S. leadership and empowers adversaries.

Beijing calls the accusations slander and demands cooperation. The U.S. sees espionage. Distillation occupies a legal gray zone that is rapidly turning black. Terms of service have been violated, but court tests are pending. The cable pushes that boundary.

Here is the core tension: China builds AI capability without the grinding investment. The U.S. funds the frontier. Outputs are siphoned. The cycle repeats. The Trump team vows to break it. Allies are looped in. The technology rivalry has gone fully diplomatic. Watch the May summit. The stakes: AI supremacy.

Source: Webpronews