The U.S. State Department has ordered a global push to bring attention to what it says are widespread efforts by Chinese companies, including AI startup DeepSeek, to steal intellectual property from U.S. artificial intelligence labs. OpenAI warned U.S. lawmakers that Chinese artificial intelligence startup DeepSeek was targeting the ChatGPT maker and the nation's leading AI companies to replicate models and use them for its own training. This isn't corporate espionage. It's strategic warfare disguised as technology transfer.
Two days earlier, the White House Office of Science and Technology Policy published a memo accusing Chinese entities of running "deliberate, industrial-scale campaigns" to distill American frontier AI systems. The diplomatic cable, per Reuters, instructed embassy staff to speak to their foreign counterparts about "concerns over adversaries' extraction and distillation" of U.S. models, naming DeepSeek alongside Moonshot AI and MiniMax.
The diplomatic cable and the V4 launch both come just weeks before President Trump is scheduled to visit Chinese President Xi Jinping in Beijing for a summit expected to cover semiconductor export controls and IP disputes.
The Distillation Machine
The allegations aren't vague corporate accusations. They're specific, documented, and devastating. Anthropic estimated that the three Chinese firms collectively generated over 16 million exchanges with Claude from around 24,000 fraudulently created accounts. Of the three firms, Anthropic found that MiniMax drove the most traffic, with over 13 million exchanges.
But distillation can also be used for illicit purposes: competitors can use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently.
DeepSeek's operation was surgical. In one notable technique, their prompts asked Claude to imagine and articulate the internal reasoning behind a completed response and write it out step by step, effectively generating chain-of-thought training data at scale. We also observed tasks in which Claude was used to generate censorship-safe alternatives to politically sensitive queries like questions about dissidents, party leaders, or authoritarianism, likely in order to train DeepSeek's own models to steer conversations away from censored topics.
That's not research collaboration. It's intellectual property strip mining.
The Scale Reveals Intent
To circumvent this, labs use commercial proxy services which resell access to Claude and other frontier AI models at scale. These services run what we call "hydra cluster" architectures: sprawling networks of fraudulent accounts that distribute traffic across our API as well as third-party cloud platforms.
This infrastructure doesn't emerge overnight. It requires planning, funding, and coordination that suggests state backing rather than entrepreneurial opportunism. In one case, a single proxy setup allegedly controlled more than 20,000 fraudulent accounts at once.
The Chinese Embassy's response confirms the stakes. "The allegations that Chinese entities are stealing American AI intellectual property are groundless and are deliberate attacks on China's development and progress in the AI industry". When diplomatic channels deploy blanket denials rather than specific rebuttals, they're protecting something valuable.
Why Export Controls Failed
These distillation campaigns expose the fundamental flaw in America's export control strategy. Washington has focused on restricting hardware access while AI capabilities increasingly live in software. Distillation is the process of training smaller AI models using output from larger, more expensive ones as part of an effort to lower the costs of training a powerful new AI tool.
Anthropic said that by exposing the distillation attempts, it demonstrates the effectiveness of export controls and shows that cutting-edge model development cannot be sustained alone through innovation without access to advanced chips. "In reality, these advancements depend in significant part on capabilities extracted from American models, and executing this extraction at scale requires access to advanced chips".
But the argument cuts both ways. If Chinese labs can achieve competitive performance through distillation, chip restrictions become less relevant. DeepSeek's timing reinforces this reality: DeepSeek on Friday released a preview of its V4 large language model, the Hangzhou-based startup's most powerful to date, with 1.6 trillion parameters and a 1 million token context window. The model is the first major frontier release optimized for Huawei's Ascend AI processors rather than Nvidia hardware.
The Talent Pipeline Problem
The IP theft allegations mask a deeper strategic vulnerability: America's dependence on Chinese talent. This methodology revealed that China produced 47 percent of the top AI talent in 2022, far surpassing the United States (ranked second), which accounted for 18 percent of the top AI talent.
A 2025 analysis by Nature found that over 30% of AI researchers publishing in top-tier U.S. journals and conferences are of Chinese origin, either born in China or educated at Chinese universities. Many of these researchers hold key positions at leading American tech firms such as Google, Meta, and NVIDIA, as well as at elite universities like MIT, Stanford, and Carnegie Mellon.
The tightening of H-1B visa regulations has led to a sharp decline in approval rates, from 46.1 percent in fiscal year 2021 to just 14.6 percent in fiscal year 2024, with Indian applicants securing 70 percent of these visas (in 2021).
America is simultaneously accusing China of stealing its AI technology while blocking Chinese talent from contributing to American AI development. This contradiction undermines both strategies.
What Companies Must Do Now
The DeepSeek controversy isn't tomorrow's threat. It's today's business reality. AI model theft through distillation represents a new category of competitive risk that traditional cybersecurity frameworks don't address.
The message for business leaders is clear: AI IP theft is not a future concern – it is a present compliance crisis. Rather than waiting for AI copyright and IP laws to come into effect, enterprises should start building their own policies now. The companies that establish robust compliance frameworks today will have competitive advantages tomorrow, while those that ignore these risks face potentially catastrophic legal and financial consequences.
Immediate Defense Measures
API Access Controls: Organizations need continuous monitoring, least-privilege access, strong identity and key management, model and data lineage controls, anomaly detection, and active threat-hunting to ensure adversaries cannot access, repurpose, or exfiltrate data and models from systems.
Behavioral Analytics: Signature-based systems fail against zero-day exploits, polymorphic malware, and novel attack techniques. Behavioral analytics establish baselines for normal user behavior, then find anomalies indicating potential theft operations.
Supply Chain Verification: Strengthen end-use and end-user verification. Enhanced diligence may be warranted where customers lack a legitimate business history, technical capability, or infrastructure consistent with advanced AI chip use.
The Strategic Competition Reality
At its core, this is not just a debate about intellectual property or cybersecurity. It is about strategic competition, and whether free societies will define the AI century or cede it to authoritarian regimes that view knowledge as power and power as control. AI is no longer just a tool. We are entering the era of agentic AI, systems capable of generating new knowledge, accelerating scientific discovery, and shaping human decision-making at scale. Whoever leads in this domain will write the rules of the emerging international order.
The DeepSeek allegations crystallize this competition. Beijing understands this. That is why China is not merely stealing technology opportunistically, it is waging a systematic campaign to siphon intellectual property, circumvent export controls, and exploit legal and technical gaps to build competitive AI ecosystems. To be effective, export controls must be paired with offensive measures, targeting theft, disrupting circumvention, and degrading stolen IP. Without this shift, current policies risk becoming self-imposed constraints while Beijing accelerates unchecked.
The State Department's global warning represents Washington's acknowledgment that the AI competition has moved beyond trade policy into national security territory. Companies operating in this environment must understand they're not just competing for market share. They're defending technological sovereignty.
The next phase of this competition will determine whether democratic values or authoritarian control shapes the AI century. DeepSeek's distillation campaigns prove that phase has already begun.