America’s NIMBY Problem with China: Open Models, Closed Minds

Less than a month ago, White House officials sounded the alarm: China was catching up to America in the AI race. According to Michael Kratsios, Director of the Office of Science and Technology, a Chinese AI company set up a “large-scale, covert” operation to steal “proprietary U.S. technology from Anthropic, a frontier U.S. AI lab.” As Kratsios explained, Moonshot AI achieved this capability in part through sophisticated distillation attacks, in which an actor sends rapid queries to an AI model and uses its responses to train a separate, cheaper model that mimics the original’s behavior. It’s an effective way to boost a model’s capabilities without building it from scratch.  

Moonshot’s distillation of Anthropic’s model was illicit. It fits the decades-long pattern of Chinese Communist Party (CCP)-backed companies siphoning American innovation, to the tune of billions of dollars per year. In this case, it paid off for Beijing again. Moonshot’s new model, Kimi K3, proved more capable than Anthropic’s Opus 4.8 at a fraction of the cost.  

At stake is the durability of America’s AI leadership. If free PRC models can compete with expensive U.S. models on most tasks, Beijing could endanger the competitiveness of America’s AI giants and imperil the economic growth their innovation has spurred. Compounding these risks is the ubiquity of Chinese models in the U.S. economy and evidence of a pro-Beijing political bias that can leave American businesses vulnerable to remote exploitation. Multiple administration officials signaled pending action. 

Two days later, CCP-controlled AI companies received a lifeline — not from Beijing, but from the U.S. private sector. A coalition of more than 200 AI and tech companies signed an open letter defending the very PRC models in the administration’s crosshairs. The letter never mentioned China. It instead focused on open-weight models, “AI models that anyone can download, inspect, modify, and run on their own infrastructure,” and on their role in making “advanced AI more accessible, adaptable, and widely available.”  

Regarding Moonshot’s operation targeting Anthropic, the letter called for “targeted legal and commercial frameworks” instead of “sweeping restrictions.” Jensen Huang, CEO of Nvidia, was more specific in an interview with Axios: treat distillation as a terms-of-use violation and settle the matter in court.  

That strategy ignores the scope of Beijing’s support for these companies. One-off court cases are no match for the industrial scale of Xi Jinping’s AI agenda. The goal should be to deter future attacks, not merely punish past offenses. Doing so, however, could compromise the viability of thousands of small businesses in America that rely on free or low-cost AI models. It could also complicate the interests of chip designers seeking to expand their customer base. For now, the open letter seems to have had its intended effect: the Trump administration has refrained from responding to China’s distillation of American AI models. 

NIMBYism and China 

The United States has a problem with the People’s Republic of China (PRC). Americans are familiar with NIMBY (“not in my backyard”) politics in their own communities. Replace “community safety and property values” with “profits and shareholders,” and the China connection comes into focus. Instead of locals opposing controversial construction projects, the opposition often comes from powerful corporations that block national security actions that would reduce revenue or hamper competitiveness. It is an unintended consequence of the deep economic entanglement between the United States and the PRC.  

At times, this dynamic plays out in particularly egregious ways. In 2020, firms like Nike and Coca-Cola — staples of American culture and soft power — reportedly lobbied against the Uyghur Forced Labor Prevention Act, which would have prohibited the importation of Chinese goods produced by slave labor. Some companies simultaneously claimed that they had no slave labor exposure, and that proving it would be difficult. The response to slave labor shouldn’t be watering down enforcement, but avoiding complicity.  

This moral agnosticism risks endangering Americans. Consider Apple’s current campaign to incorporate memory chips from CXMT, a PRC company the Pentagon recently designated as a “Chinese military company.” The material risks are similar to those posed by Huawei. President Trump’s first administration determined that allowing Huawei into American and allied telecommunications networks posed critical threats to civilian and military infrastructure, including the readiness of our nuclear arsenal. The only acceptable response was to block its inclusion in infrastructure build-out, rip and replace its legacy components from the U.S. market, and starve Huawei of access to American-made chips and chip-making tools. Successive administrations have enforced these measures imperfectly, at times haphazardly. Even so, these efforts have given the United States an edge over China at the frontier of AI innovation. Distillation attacks are among several workarounds Beijing has pursued to close the gap, underscoring the urgency of addressing the “Kimi moment.” 

Three responses are especially critical. First, comprehensively enforce export controls on AI chips to starve PRC model developers of compute. These measures should include walling off Chinese AI labs from American cloud services, a back-door method for Beijing-backed companies to access advanced chips. The Department of Commerce should also close off loopholes that permit subsidiaries of sanctioned PRC companies to access U.S. compute. Second, punish Chinese actors who engage in illicit distillation attacks. The objective should be to develop a suite of options, including sanctions and export controls, to impose sufficient financial pain to outweigh the technological gains of distillation. Of course, tracking distillation attacks in real time is difficult. For that reason, the Trump administration should consider a final proposal: banning PRC open-weight models. If Republicans and Democrats previously believed that building the next generation of telecommunications infrastructure on CCP-controlled equipment and software posed unacceptable national security risks, then both parties should agree now that American innovation in the AI age should rely on trusted code and secure data.  

Many signatories to the open letter oppose some or all of these policies. They support a hawkish approach to Beijing, at least in theory. Some have argued that America must win the AI race. Doing that, however, will come at a cost. It is impossible to maximize a strategy of economic competition premised on technological exposure to an adversary without incurring significant risk. This temptation to ignore basic trade-offs has bedeviled American foreign policy since the founding.  

Tempering the Hamiltonian Impulse 

Commerce and trade have defined American foreign policy since our nation’s inception. America’s geographic isolation was a boon to security, but a challenge to prosperity. Without access to overseas markets, the new nation would struggle to survive. As Walter Russell Mead has observed in Special Providence: American Foreign Policy and How It Changed the World, “It was an interruption of trade, rather than the loss of territory to rivals, that would most worry American foreign policy intellectuals through the first 150 years of national independence.”  

Despite this urgency, America’s early commercial relations were infused with optimism. The zero-sum, balance-of-power politics that had cursed the European continent with endless wars had no bearing on trade. “In commercial transactions,” explains Mead, “it is possible to have both a satisfied buyer and a satisfied seller.” The belief that trade can be a force for peace has defined the “Hamiltonian” tradition in U.S. foreign policy. For the ethos of material prosperity and innovation that marks America, we owe our thanks to the Hamiltonians. 

Of course, virtue left untended can decay into vice. As far back as the 1840s, American exporters joined their European counterparts in seeking to pry open China’s market for opium. What was good business for America was an economic and social disaster for China. The opium trade inverted the Qing Empire’s balance of payments and depleted its reserves. When the emperor insisted that Britain stop exporting opium, London responded with war.  

Mead captures a similar dynamic among American exporters at the time who “favored trade without regard for moral or social considerations,” and merchants “whose love of profit recognized no limits.” It would be madness to knowingly and willfully reduce profitability. And — importantly — Hamiltonians would also find it unthinkable to economically punish a trading partner with sanctions, tariffs, or export controls for moral reasons. America is an exemplar, not a policeman. Moreover, trade can shape and transform values and politics.  

Washington’s decision to admit China into the World Trade Organization (WTO) at the turn of the century captures this conviction most vividly. As politicians and corporations argued then, wealth creation would change demographics. A middle class would coalesce and demand increasingly more political liberties to protect property. It was a defensible argument, one many in Washington believed.  

Opening the economy to China certainly made Americans wealthier. In the initial years after China’s WTO accession, the U.S. manufacturing price index fell by nearly 8 percent. By some measures, prices in America would have been over twenty-five percent higher had U.S. imports from China remained at pre-WTO levels. It undoubtedly boosted revenues for multinational companies and cut costs for small businesses. Of course, China has not changed. The CCP has grown more draconian, not less. Beijing has leveraged its prosperity to change the geopolitical status quo. It is not satisfied to live peaceably with America, particularly, or with democracy generally. For Beijing, winning the AI race is a means to extend the CCP’s single-party dictatorship in China into the twenty-first century.  

Americans need to face the tension at the heart of the China competition and the AI race: economic openness and global interconnectivity have made the United States the wealthiest, most powerful nation in human history, but they have also left the nation’s innovations and democratic system vulnerable to Beijing’s predations. What worked initially is now hurting America. The debate about CCP AI models is broader than whether or how to regulate open-weight models. It is a matter of prudently weighing prosperity against safety. 

Michael Sobolik is a senior fellow at Hudson Institute and author of Countering China’s Great Game: A Strategy for American Dominance (Naval Institute Press, 2024). Follow him on X @michaelsobolik.