While Trump Meets Xi, America's AI Security Is Fracturing at Home: Stolen Models, Smuggled Chips, and a Spy Agency Fight Over Who Controls the Tech

The Summit Optics Don't Match the Ground Reality

Trump and Xi are expected to discuss AI 'risks' at the Beijing summit. Sounds responsible. The problem: one side is negotiating in good faith and the other is running 24,000 fake accounts to steal the other's technology.

According to reporting by the NY Post, Anthropic detected 24,000 fraudulent accounts that generated more than 16 million exchanges with its Claude chatbot. The company says leading Chinese labs were behind it — a process called 'distillation,' where China trains smaller, cheaper AI models on the output of expensive American ones. Effectively: theft with extra steps.

What 'Distillation' Actually Means

A leaked internal memo from Michael Kratsios, director of the White House Office of Science and Technology Policy, put it plainly: foreign entities — principally China — are engaged in 'deliberate, industrial-scale campaigns to distill US frontier AI systems.'

The danger extends beyond intellectual property theft. Anthropic warned that 'distilled' apps carry none of the original safety guardrails. These are AI models stripped of protections against helping develop bioweapons or launch cyberattacks. China isn't just stealing capability — it's stripping out the kill switch.

The Chip Smuggling Ring Nobody Is Talking About Enough

While the summit dominates headlines, federal prosecutors have been quietly busting an international smuggling operation. According to the NY Post, indictments describe servers packed with 'billions of dollars' worth of restricted Nvidia chips shipped through front companies in Southeast Asia, then rerouted to Hong Kong and mainland China.

One defendant was caught on surveillance video using a hair dryer to swap serial number stickers between chips — to disguise restricted hardware as compliant. These are the chips needed to train cutting-edge AI models. The U.S. restricted their sale to China for a reason. China found a way around it anyway.

Beijing Is Also Playing Defense

China isn't just stealing — it's locking down its own AI ecosystem. The CCP blocked a $2 billion Meta acquisition of a Chinese AI startup called Manus. Then, Chinese authorities prevented Manus' two founders from leaving the country.

China blocked a deal that would have brought American investment into China — because it doesn't want U.S. hands anywhere near its domestic AI assets. Xi wants to lead what he calls an 'epoch-defining technology,' and he's not interested in sharing or selling.

Washington Can't Even Agree on Who Runs AI Security

The U.S. government is having a bureaucratic turf war over who gets to evaluate AI models. According to Defense One, the Trump administration is split over whether the Commerce Department or the intelligence community should oversee AI model evaluations. Commerce officials are actively pushing back against a White House proposal to house an AI evaluation center within the intelligence community.

Meanwhile, the NSA — the nation's premier code-breaking and cyber agency — has been quietly testing Mythos, a major Anthropic model held back from public release because of its advanced hacking capabilities, according to multiple sources familiar with the matter cited by Defense One.

Rep. Jim Himes (D-CT), top Democrat on the House Intelligence Committee, said Tuesday that spy agencies should have early access to advanced AI models. But instead of acting on it, the administration is debating org charts.

The DOD vs. Anthropic Mess

The Department of Defense designated Anthropic — the company whose models the NSA is actively using — a supply chain risk earlier this year after Anthropic refused to meet certain Pentagon demands. Anthropic filed legal complaints in response.

Himes told the Politico Security Summit that the legal spat hasn't set back the intelligence community 'in the near term' — an indication that longer-term damage remains unclear.

Coverage and Reality

Left-leaning outlets like the NYT are framing the summit AI discussion as a diplomatic opportunity — two superpowers responsibly engaging on technology risk. Center-right coverage gets closer to the reality of Chinese theft but sometimes overstates how limited China's AI capabilities are. China has real capabilities, real investment, and a real long-term strategy.

Little coverage connects the domestic policy dysfunction — the Commerce vs. intelligence community turf war, the DOD-Anthropic legal fight — to the external threat. The government is fractured on multiple fronts at once.

The Situation

China is stealing American AI at industrial scale. It's smuggling the chips to run that AI. And it's blocking any American access to its own ecosystem.

The U.S. response is: a diplomatic photo op in Beijing and an internal argument about which government department gets to be in charge.

While Trump and Xi shake hands, China is using a hair dryer to peel stickers off stolen chips.