Commerce Dept. gains power to sanction foreign AI researchers over model theft

What it does

This bill directs the Commerce Department to identify foreign entities and individuals from China, Russia, and other adversarial countries that are stealing the technical blueprints of U.S. AI models through unauthorized access and fraudulent account schemes. Once identified, these entities face sanctions—asset freezes and transaction bans—and are added to an export control blacklist. The bill protects legitimate AI research conducted with permission and carves out humanitarian exceptions to sanctions.

Why we flagged it

The bill's operative mechanism is a national security assessment followed by export controls (Entity List additions) and economic sanctions (asset freezes) against foreign entities engaged in AI model theft. It is fundamentally a trade-control and sanctions statute, not a research or innovation bill despite its framing around 'national security interests.'

What the text implies

• The bill grants private AI companies (OpenAI, Anthropic, Meta, etc.) significant gatekeeping power over what counts as 'unauthorized extraction.' Their terms of service and contractual definitions become de facto national security policy, with minimal public oversight or appeal mechanism.
• The definition of 'model extraction attack' relies on inference from 'totality of circumstances'—volume, timing, pattern, coordination—which is inherently subjective and could capture legitimate security research, competitive benchmarking, or academic study if conducted at scale.

The full analysis lists 5 implications of this text.

Who stands to gain

U.S. closed-source AI model owners (OpenAI, Anthropic, Google DeepMind, Meta, Microsoft, etc.); U.S. technology and semiconductor companies (AMD, HPE, Cisco, Cognizant) that supply AI infrastructu; U.S. defense and intelligence contractors that rely on advanced AI capabilities