Anthropic claimed Chinese AI companies used the Claude AI model without authorisation to improve their own AI performance, TechCrunch reported on Sunday.

Anthropic alleged that Chinese AI startups DeepSeek, Moonshot AI and MiniMax created more than 24,000 fake accounts and interacted with Claude more than 16 million times, using a distillation technique to imitate Claude. They are known to have focused on replicating Claude’s agent reasoning, tool use and coding functions.

Distillation can be summarised as developing an AI model by using a teacher model and a student model. The teacher model typically draws heavily on commercial LLMs such as those from OpenAI. It generates words that are likely to come next and uses them to train the student model. This can quickly transfer the teacher model’s knowledge and predictive capabilities to the student model.

In the United States, President Donald Trump announced he would ease AI chip exports, but OpenAI CEO Sam Altman (샘 알트먼) criticised it as a “disastrous mistake”.

Against this backdrop, Anthropic warned that AI chip exports would further fuel distillation attacks, weakening the competitiveness of the U.S. AI industry and threatening national security. Some argue that AI chip exports raise the profitability of U.S. companies and provide an opportunity to manage Chinese AI systems.