News / US / cnbc
Anthropic Alleges Illicit AI Model Distillation by Chinese Tech Firms
cnbc · 2026-09-03
ANTHR.FG, BABA, BABA, GOOGL, AMZN, MSFT, CRM
Anthropic claims foreign entities are using fraudulent accounts and stolen data to illegally distill its Claude models for competitive advantage.
What Happened
Intellectual Property Theft Allegations: Anthropic has accused several Chinese AI labs, including Moonshot AI and Alibaba, of illegally distilling its Claude models to develop competing technology. The company claims these actors utilize fraudulent accounts and stolen credentials to bypass security controls, effectively stealing proprietary research to create cheaper, copycat versions of their frontier AI.
The Mechanics of Distillation: While model distillation can be a legitimate practice, Anthropic’s threat intelligence team describes an illicit ecosystem where bad actors scrape model outputs at scale to train their own 'student' models. By creating hundreds of thousands of fake accounts, these entities evade detection and circumvent the safety guardrails that Anthropic and other major AI labs have implemented.
National Security and Regulatory Risks: Beyond commercial competition, Anthropic highlights significant national security concerns regarding the potential misuse of its technology by foreign adversaries. The company warns that these distilled models lack the necessary safety protocols, creating risks related to surveillance and the potential development of biological weapons, which has drawn attention from U.S. policymakers.
Industry-Wide Cybersecurity Challenges: The issue of unauthorized model access extends beyond China to include other sanctioned regions, with bad actors leveraging dark web marketplaces to facilitate their operations. As major AI labs race to expand accessibility, cybersecurity experts note that distinguishing between legitimate high-volume users and malicious distillation campaigns remains a complex, ongoing challenge for the industry.