Technology
Anthropic alleges foreign competitors are using illicit networks to scrape and copy AI models
The AI developer claims labs in China and sanctioned states are using dark web resources and fraudulent accounts to train rival models without authorization.
The short version
- Anthropic reported that foreign entities, primarily Chinese AI labs including Moonshot AI, Alibaba, DeepSeek, and MiniMax, are improperly scraping outputs from its Claude models to build competing technology.
- Company officials and cybersecurity experts say bad actors in countries such as China, Iran, Russia, and North Korea are bypassing geographic restrictions using stolen credit cards and dark web accounts.
- Anthropic raised concerns over intellectual property theft and national security risks, arguing that copycat models often strip out safety safeguards.
- The targeted foreign firms have not publicly responded, while the U.S. government has signaled plans to explore accountability measures.
Key facts
- Anthropic's threat intelligence team stated that Chinese AI labs—including Moonshot AI, DeepSeek, MiniMax, and Alibaba—are illicitly distilling Claude models to train cheaper rival products.[CNBC]
- Bad actors circumvent access and sanctions restrictions by using the dark web and fraudulent accounts purchased with stolen payment details to query models at scale.[CNBC]
- OpenAI and Google have previously documented facing similar unauthorized distillation efforts by foreign laboratories.[CNBC]
- The Trump administration issued an April memo declaring unauthorized distillation that harms U.S. research unacceptable and vowed to pursue measures to hold foreign entities accountable.[CNBC]