They accuse of plagiarism by AI

NEW YORK (EFE).— The American company Anthropic denounced a “distillation attack campaign” against its Claude model and accused the Chinese laboratories DeepSeek, Moonshot AI and MiniMax of fraudulent techniques to replicate its capabilities.

According to the firm, it was a coordinated scheme to extract knowledge from its artificial intelligence system.

“We have identified industrial-scale campaigns by DeepSeek, Moonshot AI and MiniMax to illicitly extract Claude’s capabilities to improve their own models,” Anthropic said in a statement. The company maintains that the operation violated terms of service and regional access restrictions.

According to the San Francisco-based technology company, the three laboratories made more than 16 million interactions with Claude through some 24,000 fraudulent accounts.

These actions constitute what he described as “distillation attacks”, a practice that consists of training a less powerful model from the results that are generated by a more advanced one.

Distillation is a legitimate and widely used method in the AI ​​industry to optimize systems. However, Anthropic stressed that in this case it was “illicit” because “fraudulent accounts and proxy services” were used, hiding the real location of users to access on a large scale without being detected.

“The volume, structure, and focus of the prompts were different from normal usage patterns, reflecting deliberate capability extraction rather than legitimate use,” the company argued.

The report also indicated that DeepSeek would have sought to extract Claude’s internal reasoning to create “safe alternatives to censorship for politically sensitive queries.”

Anthropic warned that illicitly distilled models “lack necessary safeguards” to prevent “possible development of biological weapons or malicious cyber activities,” posing a “significant risk to national security.”

The company asked to strengthen protection mechanisms against this type of practices in the industry.

Related