Introduction
U.S. artificial-intelligence startup Anthropic has accused three Chinese AI companies of setting up over 24,000 fraudulent accounts with its Claude AI model to siphon data and improve their own systems.
The Accused Companies
The three companies in question are DeepSeek, Moonshot AI, and MiniMax. According to Anthropic, these companies prompted Claude more than 16 million times, extracting valuable information from Anthropic’s system to train and enhance their own products.
Distillation Tactic
Anthropic’s rival, OpenAI, had earlier this month sent a memo to House lawmakers accusing DeepSeek of using the same tactic, called distillation, to mimic OpenAI’s products. While distillation has legitimate uses, such as building smaller versions of products, it can also be exploited to build competitive products quickly and at a lower cost.
Scale of Distillation Activity
The scale of distillation activity varied among the companies. DeepSeek engaged in 150,000 interactions with Claude, whereas Moonshot and MiniMax had significantly more interactions, totaling over 15 million prompts.
Implications and Concerns
This incident raises concerns about data security and the potential for AI systems to be exploited for malicious purposes. As the use of AI technology becomes more widespread, it is essential to implement robust measures to prevent such incidents and protect intellectual property.
