Anthropic claims Chinese AI firms secretly use Claude. Is it true?

The central question raised by the allegations is whether Chinese AI companies secretly accessed US models to bridge their technology gap. A joint advisory from the National Security Agency, FBI, and Cybersecurity and Infrastructure Security Agency last month claimed Chinese AI developers extracted capabilities from US models to lower their computing, electricity, and research costs.

Anthropic specifically alleged that Chinese AI developers used fraudulent accounts and proxy networks to siphon reasoning data from its Claude model. In some instances, Anthropic further claimed that its US model was used as a covert backend for Chinese applications, suggesting a direct integration rather than just data extraction.

If these claims are accurate, Beijing's primary concern would center on data security rather than intellectual property theft or direct model copying, according to one analyst cited in the article. This perspective highlights the potential for sensitive information to be exposed or compromised through such indirect usage of foreign AI infrastructure.

The focus on data security as the main concern, rather than IP, distinguishes this situation from traditional technology transfer disputes. It implies that the risk for China is less about the replication of technology and more about the integrity and control of information flowing through or processed by these allegedly accessed US models.