February 23 (GeokHub) - Artificial intelligence company Anthropic said three Chinese AI firms improperly accessed its chatbot Claude to extract capabilities aimed at strengthening their own models, raising concerns about intellectual property protection and national security.
In a statement published Monday, Anthropic alleged that DeepSeek, Moonshot AI, and MiniMax generated more than 16 million interactions with Claude through roughly 24,000 accounts that violated the company’s usage policies and regional access restrictions.
Distillation Tactics Under Scrutiny
According to Anthropic, the companies employed a technique known as “distillation,” where a smaller or less advanced model is trained using the outputs of a more capable system. By repeatedly querying Claude and analyzing its responses, rival developers can attempt to replicate reasoning patterns and improve performance in areas such as coding, data analysis and complex task orchestration.
Anthropic said it detected one of the campaigns while it was still active. In one instance, the company noted that following the release of a new model version, activity from one of the firms shifted rapidly to capture capabilities from the latest system.
The allegations come shortly after OpenAI warned U.S. lawmakers that foreign AI firms were attempting to replicate leading American AI models for competitive advantage.
National Security and Export Controls
Anthropic argued that improperly distilled models may lack critical safeguards, potentially creating broader security risks if deployed without sufficient oversight. The company suggested that such incidents reinforce the case for maintaining export controls on advanced semiconductor chips, which limit access to the computing power needed for both large-scale model training and advanced replication efforts.
By restricting high-end chip exports, policymakers aim to reduce the ability of foreign companies to independently develop or replicate frontier AI systems at scale.
