U.S. AI companies get government muscle to block foreign model theft

What it does

This bill directs the U.S. government to identify foreign entities and countries conducting 'model extraction attacks'—unauthorized attempts to steal the technical blueprints of American AI models—and authorizes the President to impose economic sanctions and add offenders to export-control blacklists. It establishes a government assessment process, creates a public list of identified attackers, and requires the State Department and Commerce Department to coordinate with AI companies to detect and deter such theft.

Why we flagged it

The bill's core function is to protect closed-source AI models from foreign extraction through government assessment, sanctions, and export-control mechanisms. While framed as national security, the operative provisions primarily serve to enforce private IP rights and restrict foreign access to American AI technology.

What the text implies

• The 'model extraction attack' definition relies on inference from 'totality of circumstances' (volume, pattern, timing of queries), which is subjective and could capture legitimate research, competitive benchmarking, or access by foreign academics without explicit intent to steal.
• The bill empowers the President to impose IEEPA sanctions on entities of concern, but 'entity of concern' is defined circularly to include any foreign person 'conducting or attempting to conduct a model extraction attack'—the very conduct the bill aims to deter. This creates a potential feedback loop where government designation precedes evidence.

The full analysis lists 5 implications of this text.

Who stands to gain

U.S. AI model developers and closed-source AI companies (OpenAI, Anthropic, Meta, Google, Microsoft,; U.S. technology and semiconductor companies (AMD, HPE, Cisco, Cognizant) that supply AI infrastructu; U.S. cybersecurity and threat-detection firms that may contract with government for assessment and m