Staff Reports
A recent investigative report exposes a glaring national security gap: American startups are selling the same high-value AI training datasets to U.S. labs and the U.S. military while also selling them to Chinese tech giants. That means the brains we train here are being packaged and shipped to potential adversaries — all while Washington focuses mostly on chips. It’s time to stop pretending data flows are harmless and start treating them like the strategic asset they are.
What the Forbes probe found
The report shows a roughly $500 million annual data supply chain where Silicon Valley firms refine instructions and labels that teach AI to do hard tasks — coding, financial modeling, cybersecurity — and then resell those refined datasets to Chinese buyers. Names like Surge AI, Mercor, AfterQuery, and Turing pop up as U.S. suppliers that also took business from Tencent, Ant Group, Alibaba, and ByteDance. Tencent, remember, has been flagged for ties to China’s military; handing them American-made cybersecurity and self-improving AI training sets should feel like handing the keys to the kingdom.
Why AI data exports matter
People talk a lot about chips, and rightly so. But chips without top-tier training data are like engines without maps. High-quality training datasets are the next most important input after computation. When U.S. firms package expertise into off-the-shelf datasets, they give Chinese labs a shortcut past costly trial-and-error. The result: cheaper, competitive Chinese models that can match or undercut U.S. offerings. If you care about American AI leadership, this practice is not just embarrassing — it’s dangerous.
What President Trump and Commerce Secretary Howard Lutnick should do
The Department of Commerce, via the Bureau of Industry and Security, already regulates dual-use tech under the Export Administration Regulations. That authority can — and should — be expanded to cover high-value AI training data. Require licensing for dataset exports to China, block sales to firms linked to the People’s Liberation Army, and make companies with federal contracts disclose all foreign data customers. If necessary, classify advanced instruction sets as controlled dual-use items and impose penalties for noncompliance. Commerce Secretary Howard Lutnick should stop hiding behind bureaucracy and move fast; this isn’t a policy wonk exercise, it’s national defense.
Conclusion: Stop subsidizing our competitors
It’s a simple moral and strategic point: you can’t claim to be building American advantage while selling that advantage to adversaries for cash. The GOP’s tough talk on China means nothing without policy that backs it up. Tighten export controls on AI data, enforce them, and make the startups choose whose side they’re on. Otherwise we’ll keep training the machines that will one day compete with — or harm — American interests, and we’ll have nobody to blame but ourselves.
