US Agencies Accuse Six Chinese AI Labs of Large-Scale Model Distillation

U.S. cyber and intelligence agencies have publicly accused a group of Chinese artificial‑intelligence firms of conducting massive "distillation" operations against leading American models. The CISA, NSA and FBI named DeepSeek, Moonshot, Alibaba, MiniMax, StepFun and an additional lab, alleging that they have been harvesting the behavior of frontier models—including Claude, GPT, Gemini and Grok—through millions of queries and fake accounts since late 2024 .

The accusations come amid an escalating AI cold war that initially centered on hardware access, such as Nvidia H100s and Blackwell chips. While that battle continues, Washington is now focusing on the data and model‑level tactics that can bypass export controls. By extracting how models think rather than stealing raw weights, the Chinese labs aim to recreate comparable capabilities without direct hardware imports.

Details released by the agencies describe an "industrial‑scale" campaign that leverages gray‑market transfer stations to hide the origin of queries and punch through geographic restrictions. The effort reportedly involved thousands of fabricated user accounts to generate the massive query volume needed for effective distillation . Bloomberg and other outlets have highlighted Alibaba and DeepSeek as primary actors in this systematic siphoning.

Analysts warn that such behavior could erode the competitive edge of U.S. AI developers and accelerate the diffusion of advanced models abroad. Policymakers may consider tightening export controls on model outputs, enhancing monitoring of query patterns, and pursuing diplomatic pressure to curb illicit data harvesting. The next weeks will likely see further regulatory proposals and possible legal actions against the identified firms.

Related Stocks

• GOOGL
• GAP