Chinese AI labs DeepSeek, Moonshot, and MiniMax have been accused by Anthropic of conducting large-scale model distillation campaigns to replicate its proprietary systems.
Anthropic identified three Chinese firms using distillation to copy its models
The AI safety and research company Anthropic disclosed earlier this year that it detected coordinated efforts by DeepSeek, Moonshot, and MiniMax — all based in China — to recreate its language models through distillation techniques. These efforts involved querying Anthropic’s systems extensively to train smaller, comparable models without direct access to the original architecture or weights. Anthropic characterized the activity as a systematic campaign rather than isolated incidents.
The White House framed the activity as mass AI theft in an internal memo
A White House memorandum, circulated internally earlier this year, characterized the distillation efforts as evidence of widespread AI intellectual property theft by Chinese entities. The memo used Anthropic’s findings to support a broader claim about state-linked or state-tolerated efforts to bypass U.S. Export controls and replicate advanced American AI capabilities. It did not name specific officials or agencies behind the assessment.
Beijing has not publicly responded to the allegations
Chinese officials and the three named companies have not issued public denials or confirmations regarding the specific claims made by Anthropic or referenced in the White House memo. Distillation, while a known technique in machine learning, operates in a legal and ethical gray area when used to replicate proprietary models without permission. Experts note that proving such activity definitively requires access to training logs or model outputs that are rarely disclosed.
What is model distillation in AI?
Model distillation is a technique where a smaller model is trained to mimic the behavior of a larger, more complex model by learning from its outputs. While it can be used legitimately for efficiency, it can also be employed to reverse-engineer proprietary systems when direct access is restricted.
Why does the White House view this as a national security issue?
The memo suggests that circumventing U.S. AI advancements through replication could erode technological competitiveness and undermine export control strategies designed to limit adversarial access to cutting-edge AI. However, the memo does not allege direct state involvement by the three firms named.
