AI Espionage Wars: U.S. Targets China Over “Model Distillation” Allegations

The invisible arms race in artificial intelligence just hit a new fever pitch. As the U.S. government sounds the alarm over Chinese companies allegedly “distilling” and repackaging American AI models, a shadowy contest for technological supremacy is unfolding-one where lines between innovation, espionage, and competition blur at digital speed.

On Thursday, the Trump administration’s top science adviser, Michael Kratsios, issued a sharp warning: foreign entities-“principally based in China”-are orchestrating “industrial-scale” efforts to extract the capabilities of premier U.S. AI models. The memo, arriving amid escalating U.S.-China tech tensions, promises new defensive measures and punitive actions against offenders. The message is clear: the White House sees the AI race as a zero-sum contest for global influence, economic advantage, and military might.

At the heart of the dispute is “model distillation,” a technique that can legitimately train smaller AI systems by mimicking the behavior of larger, more powerful models. But U.S. companies like OpenAI and Anthropic allege that Chinese labs-including the rising startup DeepSeek-are abusing this process to shortcut years of research, copying the “brains” of leading American models at a fraction of the original cost. DeepSeek’s sudden emergence with a competitive language model last year rattled industry leaders and fueled suspicions of intellectual property theft.

Congress is responding with rare bipartisan urgency: a new bill aims to identify and sanction foreign actors who extract “key technical features” from closed-source U.S. AI systems. “Model extraction attacks are the latest frontier of Chinese economic coercion,” said Rep. Bill Huizenga, echoing a growing consensus that AI expertise is now a matter of national security.

China’s government, meanwhile, counters that these accusations are “groundless” and reflect an attempt to stifle healthy competition. Chinese officials insist their AI sector respects intellectual property and call for greater scientific exchange. Yet the reality is more tangled. In a twist, U.S. startup Anysphere recently admitted its own AI tool was built atop a model created by China’s Moonshot AI-blurring the very lines Washington seeks to defend.

Experts warn that policing “distillation” will be like searching for needles in a haystack, given the sheer volume of legitimate and illegitimate AI development worldwide. Still, enhanced information sharing among American labs and federal coordination could help stem the tide.

The AI frontier is now a geopolitical battlefield, where accusations fly and innovation is both a shield and a sword. As both superpowers scramble to protect-and poach-the building blocks of intelligence, the real risk may be an endless cycle of suspicion that stifles the very progress each side claims to champion.

WIKICROOK

• Model Distillation: Model distillation trains a smaller AI model to mimic a larger one, enabling efficient, accurate cybersecurity solutions on resource-limited devices.
• Closed: Closed describes systems or environments that are isolated from external networks, reducing risks by limiting outside access and potential security threats.
• Open: 'Open' means software or code is publicly available, allowing anyone to access, modify, or use it-including for malicious purposes.
• Model Extraction Attack: A model extraction attack involves copying an AI model by sending repeated queries and analyzing responses, risking intellectual property and data exposure.
• Intellectual Property (IP): Intellectual Property (IP) covers creations of the mind, like inventions and works of art, that have value and are protected by law.