Anthropic has confirmed that Chinese research groups are stealing its AI models, describing the situation as “theft of American IP” based on multiple rounds of evidence. The lead of Anthropic’s Frontier Red Team expressed concern that these groups are distilling knowledge from US models, which poses significant risks, particularly as the consequences of unauthorized knowledge transfer through model distillation heighten concerns about national security in the context of advanced AI techniques.

Anthropic: Anthropic is an American AI company focused on building safe and capable AI systems, with its Claude models serving as a core product line. The company is directly addressing concerns over intellectual property in this development by publicly confirming evidence of Chinese research groups distilling knowledge from its models.

AI Security: AI companies are highlighting risks of model distillation as a method for unauthorized knowledge transfer between competing research efforts.
Geopolitics: Concerns are growing over the use of advanced AI techniques in ways that could impact national security interests.