US authorities including the NSA, FBI, and CISA accuse multiple Chinese AI companies of running industrial-scale efforts to copy American “frontier” AI model capabilities. The allegations describe activity such as “distilling” American models, a process that involves extracting learned capabilities from model outputs rather than stealing the original code or weights.
Quartz reports that the agencies name six companies—DeepSeek, Alibaba, Moonshot AI, MiniMax, StepFun, and Z.AI—and say the firms extracted “billions of tokens” from U.S. frontier models since 2024. Engadget similarly describes accusations targeting DeepSeek and Moonshot, along with other Chinese AI companies, framing the matter as systematic extraction of model capabilities.
While both outlets rely on the same U.S. agency claims, they emphasize different scope details: Engadget highlights the broader list of accused firms and the concept of distillation, while Quartz focuses on the specific company lineup and the time period and scale of token extraction. Both accounts present the allegations as ongoing claims by U.S. authorities rather than verified outcomes in court.
