Anthropic Accuses Chinese AI Firms of Covertly Exploiting Claude Model to Train Their Systems

Anthropic Levels Serious Allegations

According to НВ — Техно: Anthropic has accused several Chinese artificial intelligence companies-including DeepSeek, Moonshot AI, and Alibaba-of secretly leveraging its Claude model to train their own competing AI systems. These claims stem from a detailed report by Anthropic documenting data extraction campaigns that spanned two years and involved millions of queries. According to the company, Chinese developers extracted billions of tokens not only from Claude but also from models like GPT, Gemini, and Grok.

The Chinese government may have been aware of these operations. The report highlights that Moonshot AI, creator of the chatbot Kimi, rerouted 300,000 client requests to Claude within just ten days instead of using its own model. This was facilitated by a network of 5,380 fake accounts based in Singapore and Japan. Between May and July 2026, Alibaba reportedly conducted over 151 million model distillation operations-a technique where outputs from a more powerful model train another.

Responses and Wider Implications

While Anthropic acknowledges distillation as a legitimate AI training method, it draws a clear line between open research and unauthorized replication. The report also suggests Chinese developers employed third-party aggregators, cloud providers, and proxy services to bypass restrictions on accessing US-based AI systems. In response, US intelligence agencies-including the Cybersecurity and Infrastructure Security Agency (CISA), the National Security Agency (NSA), and the Federal Bureau of Investigation (FBI)-issued a joint alert urging American companies to strengthen their cybersecurity defenses.

Wang Zebin, an investment manager at JG Investment, expressed skepticism about verifying Anthropic’s claims independently, stating:

“The report lacks sufficient concrete examples to independently validate how these companies obtained and utilized model outputs.”

In reaction to the allegations, China’s cyberspace administration surveyed seven domestic AI developers, focusing particularly on DeepSeek and Moonshot AI. However, neither company responded to requests for comment from the South China Morning Post.

Amid these developments, NSA, FBI, and CISA have urged AI developers to:

• enhance detection systems for such unauthorized activities;
• implement targeted countermeasures;
• share intelligence and best practices.

Anthropic has vowed to put a stop to unauthorized exploitation of its technology and to bolster the security of Claude.

This controversy highlights increasing concerns over cyber threats and intellectual property misuse in the rapidly evolving global AI landscape. Anthropic’s accusations may trigger tighter regulations surrounding AI, especially regarding intellectual property protections. The responses from Chinese developers and government authorities could significantly influence future US-China relations in technology and innovation sectors.

The allegations made by Anthropic have sparked significant interest, especially in light of ongoing investigations. As reported, Chinese regulators are probing DeepSeek and Moonshot AI for potential user data breaches, which raises questions about the broader implications of AI practices in the region. Understanding these developments can provide deeper insights into the challenges faced by AI companies globally.