Congress wants Chinese AI firms prosecuted under federal law for what it’s calling industrial espionage, and your export-control playbook needs updating before the Trump-Xi meeting next month.
The tactic at issue: adversarial distillation attacks, where Chinese firms query US frontier models at scale to extract capabilities and train competing systems without paying for access or compute. A congressional committee is asking the State Department to assess whether those attacks violate the Economic Espionage Act and the Computer Fraud and Abuse Act. The committee wants to go further: officially classify adversarial distillation as a controlled technology transfer, making unauthorized model access a sanctionable offense.
The committee’s report is direct: current penalties are treated as “a tolerable cost of doing business.” That changes if distillation gets categorized alongside export-controlled chips.
Chris McGuire at the Council on Foreign Relations told the Financial Times that Chinese AI firms rely on distillation attacks to offset deficits in AI computing power, replicating US model capabilities without the hardware. The complication is Trump’s own export concessions: his deal allowing Nvidia chip sales to China in exchange for a 25 percent cut was called “no sense” by experts who warned it could open the door to deeper capability transfer. That’s a problem if some of those concessions need reversing.
Beijing’s response through embassy spokesperson Liu Pengyu: “pure slander.”
For any founder running a frontier model or building on top of one, the question Monday morning is whether your API terms-of-service hold up as a legal instrument, or whether federal classification of distillation as tech theft finally gives them teeth.
Nathan Zakhary