Policy
U.S. AI Export Controls Miss Critical Data Vulnerability
While America restricts chips and models to China, advanced training datasets remain largely unprotected—a gap that could accelerate Beijing's catch-up.
Key takeaways
- U.S. export controls focus on AI chips and models but largely ignore advanced training data, creating a strategic vulnerability.
- Leading U.S. and Chinese AI models have reached near-parity in performance over the past year, with China closing the gap rapidly.
- Chinese labs purchasing U.S. training datasets gain insight into American technical approaches and world-leading data capabilities.
- Former DHS Secretary Chad Wolf recommends treating advanced training data as a strategically significant capability requiring export safeguards.
- China generates twice the electricity of the U.S. and has cultivated half the world's AI engineering talent, providing structural advantages in the technology race.
The overlooked weakness in America's AI defenses
The United States has erected barriers around semiconductor exports and frontier AI models to maintain its technological edge over China. But a critical component of AI development—advanced training data—remains largely unprotected, according to former acting Homeland Security Secretary Chad Wolf.
Writing in The National Interest, Wolf argues that while China has made dramatic progress in AI capabilities over the past year, U.S. export controls focus narrowly on chips and algorithms while ignoring the third pillar of AI development: the specialized datasets that train cutting-edge models.
Why it matters
Chinese AI labs purchasing advanced U.S. training data gain more than raw information—they obtain insight into the technical challenges American companies are solving and access to world-leading data preparation capabilities. This creates an alternative pathway for China to close the technology gap even as the U.S. restricts other AI inputs. With leading U.S. and Chinese models now approaching performance parity, the window to establish safeguards is narrowing.
China's accelerating AI progress
The competitive landscape has shifted rapidly. Just one year ago, the United States held a clear advantage in frontier AI capabilities. In the past four months alone, major labs in both countries have released new models that demonstrate near-equal raw performance, according to Wolf's analysis.
China brings substantial structural advantages to the competition: it generates approximately twice the electricity of the United States, achieves significant economies of scale in manufacturing, and has cultivated half the world's AI engineering talent through widespread STEM education. The Chinese government can also direct capital, energy, and infrastructure toward AI priorities with speed that democratic systems cannot match.
Beijing is integrating AI across military doctrine, domestic surveillance systems, and global technology exports. The Chinese Communist Party has deployed AI-powered tools in cyberattacks like the Volt Typhoon campaign and uses the technology to track citizens' movements, communications, and online behavior.
The data dimension
Wolf emphasizes that advanced AI rests on three core ingredients: computational chips, algorithmic architecture, and training data. Current U.S. policy addresses the first two while leaving the third exposed.
When Chinese labs purchase advanced training datasets or specialized data-development services from U.S. companies, they gain access to capabilities that would otherwise require years to develop independently. This transfer accelerates China's ability to compete at the frontier of AI development.
Policy recommendations
The former DHS secretary calls for the Trump administration to treat advanced training datasets and data-development services as strategically significant AI capabilities subject to the same export scrutiny as compute resources and frontier models. He recommends targeted safeguards preventing U.S. companies from transferring these capabilities to Chinese military-linked entities and frontier AI laboratories.
Wolf frames the challenge in stark terms: if Beijing succeeds in exporting its AI technology stack and governance model globally, key networks and standards could be controlled from Beijing rather than shaped by democratic societies. He argues the window for establishing a rules-based AI order is measured in years, not decades.
The analysis was first reported by The National Interest and authored by Chad Wolf, who served as acting Secretary of Homeland Security during President Trump's first term and now serves as vice chairman at American Global Strategies.
