Washington’s anti-China tech strategy is built on a dangerous illusion

When Treasury Secretary Scott Bessent said this week that the United States was finding American “watermarks” embedded deep inside Chinese artificial intelligence models, he signaled a potential shift in geopolitical statecraft.For four years, Washington treated the technology competition with Beijing as a hardware fight — a battle over physical silicon, lithography equipment, and advanced semiconductors. That era is over. The front line has shifted from physical manufacturing to intangible software.Just as Washington historically deployed tariffs to protect domestic manufacturing from subsidized foreign goods, the Treasury’s move signals the arrival of digital tariffs, using economic sanctions to protect frontier American AI from systematic intellectual property extraction. The trigger for this escalation is not just technical — it is economic. The global debut of Chinese open-weight (publicly available) models, most notably Moonshot AI’s Kimi K3, sent shockwaves through Wall Street. By offering frontier-level reasoning capabilities at fractions of a cent per query, substantially undercutting American proprietary models, Beijing’s tech ecosystem is initiating a classic predatory dumping offensive, and the market reacted.BINDERS WON’T STOP BULLETS: WHEN MILITARY AI CAN’T SAY ‘NO’American enterprise technology officers, eager to cut soaring operational budgets, began routing in volume toward cheap foreign open weights. But what looks like a commercial bargain on a corporate balance sheet, in reality, is economic warfare.We do not need to speculate on how this playbook unfolds — we have already lived through it. It was called Huawei/ZTE. Two decades ago, state-subsidized Chinese telecommunications firms flooded global markets with hardware priced at unsustainable discounts, undercutting Western competitors like Ericsson, Nokia, and Lucent. Rural American carriers and municipal networks rushed to install the cheap infrastructure. Only after the state-backed hardware was deeply embedded into the cellular backbone of Western nations did the national security fallout become undeniably apparent. The result was a multibillion-dollar federal “Rip and Replace” program: a chaotic, expensive effort to physically tear gear out of cell towers.By treating subsidized open-weight models as harmless software discounts, enterprise leaders are making the same exact strategic miscalculation. The difference is that while replacing physical cell towers took a decade, purging backdoored synthetic cognition, with its downstream workflows and automated decisions, from an enterprise software pipeline is functionally impossible, making the breach an unprecedented, multidimensional security nightmare.The mechanism driving this extraction is known as “adversarial distillation,” a process where an attacker issues tens of millions of automated, programmatically routed application programming interface queries to top American models like Claude or GPT. Done legitimately, distillation is a standard engineering tool for building lightweight applications. Done at an industrial scale through automated queries and networks of fraudulent proxy accounts, it becomes industrial-scale digital theft, easily weaponized.By capturing the unredacted reasoning patterns and output structures of U.S. systems, foreign developers train secondary models at a fraction of the cost. Disclosures from Anthropic and OpenAI confirm that Chinese labs ran multimillion-query campaigns through tens of thousands of fraudulent proxy accounts, creating an unprecedented national security breach across major American cloud networks.Washington is beginning to formalize its response. The House Committee on Foreign Affairs unanimously advanced the Deterring American AI Model Theft Act (H.R. 8283), while White House Memorandum NSTM-4 designated adversarial distillation as a national security threat. Together, these measures would reclassify covert, industrial-scale API extractions as statutory IP theft and national security espionage, authorizing asset freezes under the International Emergency Economic Powers Act, and Entity List blacklisting. Meanwhile, reports from Beijing indicate that China’s Ministry of Commerce is consulting with its domestic technology firms about imposing export controls on downloadable model weights.Yet, this points to a broader reality: Beijing’s vigorous defense of “open-source AI” was not a commitment to global access. It appears to be more of a temporary competitive strategy to clear the market, weaken American private investment, and then pull up the drawbridge. Washington has identified the watermark — it must now implement the critical strategy to act on that discovery. However, as it moves to counter the threat, it faces a structural dilemma. China operates in this technological war with relatively little domestic accountability. The Chinese Communist Party faces no Fourth Amendment privacy barriers, no public pushback over data harvesting, no judicial injunctions, and no corporate board resistance. It can direct state capital, mandate data integration, and deploy asymmetric software tools literally overnight.Washington, by contrast, operates within a constitutional framework. When the Department of Commerce or Treasury executes a blunt administrative “kill switch”, such as the abrupt 18-day global suspension of Anthropic’s frontier models, under “deemed export” regulations, it demonstrates immense regulatory power yet also reveals a profound vulnerability.Because cloud networks cannot instantly verify users’ nationalities, a single national security directive also abruptly pulls the plug on domestic hospitals, financial institutions, and defense research labs utilizing those models.A government whose primary setting for managing frontier technology is an on-off switch is not governing — it is hovering over a panic button. When we rely on classified administrative directives to restrict software access, we do not neutralize our adversary’s advantage; we compromise our own.WIKIPEDIA IS TRAINING GOOGLE AND AI TO LIE TO YOUThe challenge facing American leadership is not acknowledging that we are in an economic war — it is understanding that we are severely handicapped across the artificial intelligence landscape against an unconstrained state actor. As Treasury prepares to issue sanctions and mandatory disclosure rules around foreign counterfeit software, the policy debate must move beyond short-term trade bans. The true crisis of the synthetic era is not merely who owns the code, but what happens to citizen privacy, corporate compliance, and constitutional rights when the digital foundation of daily life becomes an irrevocable battlefield.Jacqueline Cartier is a corporate and legislative strategist focused on communications, crisis leadership, public trust, and emerging technologies that shape human behavior and decision-making. Follow her on LinkedIn.