Anthropic reports the largest Claude distillation attack, linking it to Alibaba
Another high-profile conflict is unfolding in the artificial intelligence industry. Anthropic, the developer of the Claude model, sent an official letter to the chair of the U.S. Senate Banking Committee and senior Democrat Elizabeth Warren, accusing operators linked to China's Alibaba and its AI lab Qwen of a large-scale campaign to distill their model.
According to my data, this incident has already been called the largest known distillation attack. From April 22 to June 5, attackers used nearly 25,000 fake accounts to generate over 28.8 million interactions with Claude. This is not just a violation of the service's terms of use, but a systematic attempt to replicate the capabilities of a cutting-edge model without the colossal costs of training it.
Anthropic emphasizes that the campaign targeted Claude's key capabilities: agentic tasks, software development, and long-term planning. The company notes that Alibaba, being listed on the New York Stock Exchange and having business operations in the U.S., is accountable to American investors and regulators, making the situation particularly egregious.
It is important to understand that distillation itself is not illegal and is often used legitimately — for example, to create compact and inexpensive versions of models. However, the problem arises when competitors gain access to a cutting-edge tool through fake accounts, bypass restrictions, and use the responses to train their own systems.
Economic Implications and Call to Congress
The letter emphasizes that such actions "upend the economic logic of U.S. leadership in AI." When Chinese labs distill capabilities from American models, they reap the benefits of multi-billion dollar U.S. investments without bearing the costs and risks. Moreover, Anthropic warns that this could accelerate the development of Chinese AI systems for cyber operations, military tasks, and intelligence.
The company urges lawmakers to take several measures: expand the sharing of technical indicators between AI developers and the government, clarify antitrust rules for sharing information about attacks, tighten export controls on advanced chips and computing resources, and impose sanctions on responsible parties.
Context and Precedents
This is not the first such case for Anthropic. In February, the company already accused DeepSeek, Moonshot AI, and MiniMax of distilling Claude. According to the company, DeepSeek generated over 150,000 interactions, Moonshot AI over 3.4 million, and MiniMax around 13 million. In April, Elon Musk also admitted that xAI "partially" used OpenAI models when training Grok, highlighting the prevalence of the practice.
It is worth noting that in April, Congressman Bill Huizenga introduced a bill against the extraction of key characteristics of closed U.S. AI models by foreign adversaries, providing for export restrictions and sanctions.
My analysis: This incident is just the tip of the iceberg in the global AI race. Distillation is becoming a standard tool for countries seeking to close the technology gap. However, for Anthropic, this is not only a security issue but also a direct threat to their business model. If major players like Alibaba can legally or illegally copy Claude's behavior without licensing fees, it undermines the very economics of developing advanced models. I expect that in the coming months, we will see a tightening of both corporate protection policies and government regulation in this area.