Anthropic alleges that Alibaba illicitly extracted and copied the capabilities of its Claude AI model to enhance its own AI offerings [1].
The dispute highlights the intensifying competition and security risks surrounding large language models as companies race to dominate the global artificial intelligence market.
Anthropic said Alibaba used fake accounts to bypass security restrictions and obtain the model's capabilities [1]. The U.S. company said the incident was the largest known attack on its model to date [2].
According to reports, Alibaba allegedly sought to integrate these extracted capabilities into its own suite of AI products [1]. This method of extraction involves bypassing safety guardrails to mirror the behavior, and logic of a proprietary model.
Market reaction to the allegations was immediate. Alibaba stock fell about 2.7% [3] to a 52-week low following the announcement. The drop reflects investor concern over potential legal repercussions, and the reputational risk associated with allegations of intellectual property theft.
Alibaba has not issued a formal response to the specific claims of fraudulent account usage. The situation underscores the vulnerability of AI models to "model stealing" attacks, where competitors use API access to reverse-engineer a model's internal logic.
“Anthropic said the incident was the largest known attack on its model to date.”
This conflict illustrates the growing tension between U.S. and Chinese AI development, where the line between competitive benchmarking and intellectual property theft is increasingly blurred. If proven, the use of fraudulent accounts to mirror a competitor's model could lead to stricter API restrictions and new legal precedents regarding the ownership of AI-generated behaviors.
