US AI Giant Anthropic Accuses Chinese Rivals of Mass Data Theft
Anthropic, a leading US artificial intelligence company, has publicly accused three Chinese AI firms of engaging in large-scale intellectual property theft. The allegations centre on the illicit extraction of capabilities from Anthropic's Claude chatbot, a practice the company describes as growing in both intensity and sophistication.
Distillation Technique at the Heart of the Allegations
According to Anthropic, the Chinese companies—DeepSeek, Moonshot AI, and MiniMax—employed a method known as "distillation." This technique involves using outputs from a more advanced AI system, like Claude, to rapidly enhance the performance of a less capable model. While distillation is a common and legitimate practice within AI development for creating cost-effective versions of proprietary models, Anthropic contends that in this instance, it was used unlawfully to bypass export controls and undermine US technological dominance.
The company reported that these campaigns were executed through approximately 16 million exchanges with its Claude model, facilitated by around 24,000 fake accounts. To circumvent Anthropic's restrictions on commercial access from China, the labs allegedly routed traffic via proxy services, managing extensive networks of fraudulent accounts to mask their activities.
National Security and Industry Implications
Anthropic has raised serious concerns about the national security risks posed by such practices. The company argues that models developed through illicit distillation may lack critical safety guardrails designed to prevent misuse, such as restrictions on aiding in the development of bioweapons or enabling cyberattacks. This, Anthropic warns, could have far-reaching consequences for global security.
The issue gained prominence a year ago when DeepSeek released a low-cost generative AI model that performed comparably to top US chatbots like ChatGPT, challenging assumptions of American supremacy in the sensitive AI sector. Anthropic's accusations echo similar charges made by its rival, OpenAI, which last month told US lawmakers that Chinese entities were "free-riding" on capabilities developed by US frontier labs.
Details of the Alleged Operations
Anthropic provided specific details about the scale of the alleged theft. MiniMax was identified as running the largest operation, generating over 13 million exchanges. The campaigns reportedly focused heavily on areas where Claude excels, including coding, agentic reasoning, and tool use. By siphoning off these capabilities, the Chinese firms allegedly achieved advanced performance at a fraction of the independent development cost, effectively skirting US export controls intended to preserve American leadership in AI.
In response, Anthropic has called for a coordinated effort between industry and government to address what it describes as a challenge too vast for any single company to tackle alone. The company emphasised that the window for action is narrow, urging swift measures to curb these sophisticated theft campaigns.
This development underscores the escalating tensions in the global AI race, with intellectual property theft becoming a focal point in the strategic competition between the US and China. As AI technology continues to advance, such incidents highlight the need for robust international regulations and collaborative security frameworks to protect innovation and ensure ethical development practices.
