Anthropic Accuses Chinese AI Firms of Secretly Training Rivals with Claude Chatbot

San Francisco-based artificial intelligence firm Anthropic has accused three Chinese companies – DeepSeek, Moonshot AI and MiniMax – of engaging in “industrial-scale campaigns” to illicitly extract capabilities from its Claude chatbot. The allegations, detailed in a blog post released Monday, center on a technique known as “distillation,” where one AI model is trained on the outputs of another, more advanced system.

Anthropic claims the Chinese firms circumvented regional access restrictions and terms of service by creating approximately 24,000 fraudulent accounts and generating over 16 million exchanges with Claude. The company alleges this was done to improve their own models, specifically in areas where Claude excels – complex reasoning, coding assistance, and tool use. While distillation itself is a legitimate training method, Anthropic argues the scale and method employed by these companies constitute a violation of its terms and potentially undermine U.S. Export controls.

The accusations come amid increasing scrutiny of data sourcing practices within the AI industry. Anthropic itself recently settled a landmark copyright lawsuit for $1.5 billion with authors who alleged the company illegally downloaded their books to train its models. The company also faced accusations of unauthorized data scraping from Reddit, as reported by Decrypt. This latest dispute highlights a broader debate over the boundaries of fair use and intellectual property in the rapidly evolving AI landscape.

How the Alleged Extraction Was Carried Out

According to Anthropic, the Chinese labs bypassed geofencing and commercial restrictions on Claude’s availability in China by utilizing proxy services to resell access. These services allowed the firms to operate a “hydra cluster” of tens of thousands of accounts simultaneously, distributing requests across multiple API keys and cloud providers.

Once established, these accounts were allegedly used to script lengthy, high-volume conversations designed to elicit detailed, step-by-step responses from Claude. Anthropic contends that this created an unauthorized pipeline, effectively turning Claude into a training resource for competing models developed within China’s AI sector.

While Anthropic has not yet initiated legal action, it has taken steps to block known access points and is urging the U.S. Government to strengthen export controls on advanced chips and AI services to prevent similar incidents in the future.

A Case of Hypocrisy?

The allegations have been met with skepticism from some industry observers, who point to Anthropic’s own controversial data collection practices. Commentators have highlighted the irony of Anthropic accusing others of extracting value from proprietary systems while simultaneously facing criticism for its own methods. As one commenter noted on Reddit, “How the turn tables,” referencing a popular meme.

This situation underscores a fundamental tension within the AI industry: U.S. Firms like Anthropic and OpenAI are aggressively pursuing legal protections for their proprietary systems, even as they defend their own extensive data collection practices under the umbrella of fair use. Chinese labs, often releasing more open-source models, are seeking to close the performance gap with Western rivals by leveraging any available advantage.

The rapid advancement of DeepSeek, which launched a model last year rivaling those of ChatGPT with comparatively fewer computing resources, has already raised questions about the effectiveness of U.S. Export controls. OpenAI leveled similar charges against DeepSeek earlier this month, as reported by The Guardian. Anthropic’s latest accusations are likely to intensify calls for stricter regulations on AI technology transfer to China, while simultaneously prompting further debate about the ethical and legal boundaries of data sourcing in the AI era.

The accusations from Anthropic are occurring as the U.S. Government is already considering tighter restrictions on exporting AI chips and cloud services to China.