Training a leading AI model costs billions of dollars and requires huge, carefully prepared datasets. But a secretive and shady side of the AI industry has emerged, letting some companies avoid these costs by using a method called “distillation.”
This week, the hidden battle became public when Anthropic, the company behind the Claude chatbot, accused three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—of stealing its AI knowledge on a large scale.
Distillation works like an intellectual property theft using an API. Instead of building a model from scratch, a company asks a bigger, smarter “teacher” AI model millions of questions. It then feeds the high-quality answers into a smaller, cheaper “student” AI model. Over time, the student model learns to think like the teacher without ever paying for the expensive training process.
According to Anthropic, the three Chinese labs made over 16 million queries to Claude. They used 24,000 fake accounts to collect Claude’s advanced coding and reasoning abilities, essentially copying its intelligence without permission.
This case highlights a growing problem in AI: the race to build cutting-edge models is so expensive that some companies are resorting to secretive methods to shortcut the process, raising serious legal and ethical concerns.