Anthropic has leveled serious allegations against three prominent Chinese AI developers—DeepSeek, Moonshot AI, and MiniMax—accusing them of a coordinated campaign to “mine” its Claude model. The company claims these labs used over 24,000 fake accounts to generate 16 million exchanges, effectively siphoning Claude’s logic to bolster their own systems.
The Art of “Distillation”
The core of the dispute involves distillation, a process where a smaller or rival model is trained using the outputs of a superior “teacher” model. While developers often use this internally to create faster, cheaper versions of their own software, Anthropic characterizes these external attacks as an illicit shortcut to bypass years of R&D. OpenAI recently voiced similar concerns regarding DeepSeek’s rapid rise.
Scale of the Extraction
The data mining targeted Claude’s most advanced features, including agentic reasoning, computer vision, and complex coding:
- MiniMax: Allegedly the most aggressive, redirecting nearly half its traffic to scrape Claude during a recent launch, totaling 13 million exchanges.
- Moonshot AI: Targeted data analysis and tool use across 3.4 million exchanges.
- DeepSeek: Conducted 150,000 exchanges focused on foundational logic and navigating policy-sensitive queries.
Geopolitical and Security Implications
These revelations arrive amidst a heated U.S. debate over AI chip export controls. While the Trump administration recently moved to allow Nvidia to export H200 chips to China, Anthropic argues that the massive scale of these distillation attacks proves that Chinese labs still require high-end compute to process stolen data.
Furthermore, Anthropic warns that distillation poses a national security threat. When a model is “cloned” this way, the original safety guardrails—designed to prevent the creation of bioweapons or cyberattacks—are often stripped away. This allows powerful capabilities to proliferate without the ethical oversight built into Western frontier models.
Anthropic is now calling for a unified industry and policy response to defend against these sophisticated extraction techniques.
