Eileen Wang recently resigned as mayor of Arcadia, a leafy suburb east of Los Angeles. She agreed to plead guilty in federal court to acting as an illegal agent of the People's Republic of China. From 2020 through 2022, prosecutors say, Wang took orders from Chinese officials over WeChat, posted their propaganda on a sham news site, and once thanked her handler with the words "Thank you, leader.”
Arcadia has 56,000 residents. Beijing thought the place was worth the effort. That alone should worry anyone who assumes the Chinese Communist Party is too busy stealing fighter-jet blueprints to bother with small stuff. The Party plays a long game and a wide one. If a five-person city council in the San Gabriel Valley merits an operative, the conclusion writes itself.
Now consider where the consequential work happens. In a few dozen open-plan offices around the Bay, engineers are building systems that will run the power grid, draft legal briefs, fly drones, and rewrite the rules of war.
The talent ledger reads like a Beijing alumni newsletter. When Mark Zuckerberg launched Meta's Superintelligence Labs last summer, seven of the first eleven hires came from China. The chief scientist among them. The pattern repeats across the industry. Chinese researchers sit among the co-founders of Elon Musk's xAI, Mira Murati's Thinking Machines, and Yann LeCun's AMI Labs. Roughly half the authors at NeurIPS, the field's marquee conference, did their undergrad in China. According to Nvidia's Jensen Huang, half the world's AI researchers are Chinese, and a sizable share now work on American soil.
This isn’t a claim that every Chinese-born engineer is a spy. Most are brilliant scientists who came here because America offered something their home country didn’t.
But to believe Beijing hasn’t embedded operatives in the most consequential infrastructure ever built by human hands requires a suspension of disbelief detached from how every prior strategic technology has been contested. China has a written national strategy, signed by Xi Jinping, called Military-Civil Fusion. It explicitly directs civilian AI breakthroughs into military applications. China's AI Plus program, baked into the 15th Five-Year Plan, treats commercial labs abroad as harvestable resources. The State Department warned years ago that Beijing's plan is to acquire "the intellectual property, key research, and technological advances of the world's citizens" for military gain.
In January, former Google engineer Linwei Ding was convicted on 14 counts of economic espionage and trade-secret theft for siphoning more than 2000 pages of Google's AI infrastructure secrets, including the architecture of its Tensor Processing Unit chips, to companies linked to the Chinese state. The FBI called it the first AI-related economic espionage conviction in U.S. history. The word "first" implies a sequel.
Palantir's Alex Karp has described Chinese spying on U.S. tech firms as "a huge problem," singling out large language model developers. In other words, the CEO of America's premier defense-software company is saying that the labs racing to build superintelligence are precisely the labs Beijing is racing to penetrate. A former CIA officer recently testified to the Senate Judiciary Committee that American startups are "not competing against Chinese rivals in any normal sense" but against "the largest intelligence apparatus in the world."
The asymmetry comes into focus the moment you run it backwards. Picture an American engineer applying to lead model training at a flagship Chinese AI lab in Hangzhou. He gets a desk next to the chief scientist. He attends weekly meetings about military applications. He flies home for Thanksgiving and comes back without a tail. The thought is absurd. Beijing would never permit it. Their labs are walled gardens with guard towers.
The asymmetry has a name. It is called losing.
Washington has spent four years arguing about export controls on Nvidia chips. Meanwhile, the talent pipeline, the commanding height, has been left wide open, defended chiefly by HR diversity statements and the honor system. There’s no equivalent of a security clearance for frontier AI work. There’s no equivalent of the screening that protects nuclear secrets, even though many people, including the labs' own CEOs, insist this technology is more dangerous than nuclear weapons.
Fixing this doesn't require McCarthyism; it requires the basics. Background checks worthy of the stakes. Compartmentalized access to model weights and training infrastructure. FBI liaisons embedded in the major labs, the way they sit inside defense contractors. Real consequences for companies that look the other way. And, above all, American companies with more American-born talent leading the way. Every true American wants that. China's rulers want the exact opposite.