The US company Anthropic accuses three Chinese AI firms of massively stealing data from its Claude model.
The irony: The company itself paid $1.5 billion in September 2024 for copyright violations.
The accusation
According to Anthropic, DeepSeek, Moonshot AI, and MiniMax created over 24,000 fake accounts and interacted with Claude 16 million times – to train their own models.
The technique: "Distillation" – a large "teacher" model trains a smaller "student" model. Legal when done with your own system. Illegal when tapping into foreign systems.
China's "AI tigers" in the crosshairs
The accused companies belong to China's absolute elite and challenge US market leaders like OpenAI and Anthropic:
Startup | Best-known model | Status |
|---|---|---|
DeepSeek | DeepSeek-V3 | Shocked the industry with high efficiency at low computing power. |
MiniMax | abab | Specialized in AI characters and social interaction. |
Moonshot AI | Kimi | Known for processing extremely long contexts. |
"How the turn tables"
The reactions on the internet were consistently critical.
Why? Because of Anthropic's own history:
September 2024: $1.5 billion settlement – largest copyright settlement in US history
Accusation: 500,000 books illegally downloaded and scanned from shadow libraries
Payment: ~$3,000 per book to authors
OpenAI also made similar accusations against DeepSeek – and is meanwhile itself embroiled in 16 consolidated copyright lawsuits.
Sources: CNN, The Register, Anthropic Press Release
Get These Insights Every Morning
Join 18,000+ professionals who start their day with Asiabits. Free, every weekday, straight from Shanghai.
Subscribe Free →
