US artificial intelligence company Anthropic has accused three Chinese AI firms of engaging in industrial-scale intellectual property theft by illicitly extracting capabilities from its Claude chatbot. The accused companies, DeepSeek, Moonshot AI, and MiniMax, utilized a technique called ‘distillation’ to rapidly enhance the performance of their AI systems by leveraging outputs from Anthropic’s more powerful model.
Anthropic highlighted the growing intensity and sophistication of these alleged campaigns, emphasizing the urgent need for action to address the issue. Distillation, a common practice in AI development, enables companies to create cost-effective, compact versions of their models.
The impact of distillation became evident when DeepSeek released a low-cost generative AI model that rivaled top American chatbots, challenging assumptions of US dominance in the sector. Anthropic revealed that the Chinese firms conducted approximately 16 million exchanges with its Claude model and 24,000 fake accounts to acquire capabilities they had not autonomously developed.
By circumventing export controls on US technology, the Chinese companies may have posed national security risks, as models created through illicit distillation may lack safety guardrails to prevent misuse. Anthropic’s rival, OpenAI, also raised similar concerns to US lawmakers, accusing Chinese entities of leveraging distillation to exploit capabilities developed by US labs.
These allegations underscore the significance of safeguarding intellectual property in the AI industry and the need for enhanced measures to prevent unauthorized extraction of proprietary technologies.
Source: mint – technology