Anthropic has accused Alibaba-linked operators of a large-scale distillation attack on its AI model, Claude, involving millions of fraudulent interactions.
Anthropic has accused operators associated with Alibaba and its AI lab Qwen of orchestrating a significant distillation attack against its AI model, Claude. This allegation was outlined in a letter sent to U.S. Senate Banking Committee leaders, including Chairman Tim Scott and senior Democrat Elizabeth Warren, as reported by Reuters.
The company characterized this incident as the largest known distillation attack to date. Between April 22 and June 5, it claims that over 28.8 million interactions with Claude were generated through nearly 25,000 fraudulent accounts.
"The scale and audacity of this campaign are notable. Alibaba is publicly traded in the U.S. and must answer to American regulators and investors," the letter stated.
Anthropic explained that while distillation can be a legitimate practice for creating more efficient AI models, issues arise when competitors exploit fake accounts to bypass service restrictions and use the AI's outputs to enhance their own systems.
The alleged campaign was said to focus on Claude's abilities in various applications, including software development and strategic planning, enabling competitors to mimic a leading model's performance without incurring the costs of training it from scratch.
"When Chinese labs distill capabilities from American models, they benefit from U.S. investments without the associated costs and risks, undermining the economic foundation of U.S. leadership in AI," Anthropic emphasized.
Furthermore, the company warned that unauthorized distillation could expedite the development of Chinese AI technologies for military and intelligence purposes.
Anthropic's Requests to Congress
In light of these concerns, Anthropic has urged lawmakers to enhance information sharing regarding technical indicators and intelligence between AI developers and the U.S. government. The company is also seeking clearer antitrust regulations to facilitate information exchange about such attacks without breaching competition laws.
Additionally, Anthropic has called for stricter export controls on advanced AI chips and computational resources, along with measures to close loopholes that allow Chinese entities access to foreign data centers. They suggested implementing sanctions against those responsible for large-scale extraction of AI model capabilities.
Previous Allegations
Earlier in February, Anthropic accused several firms, including DeepSeek, Moonshot AI, and MiniMax, of generating over 16 million interactions with Claude through approximately 24,000 fake accounts. At that time, the accusations were linked to specific labs based on IP addresses and other metadata. DeepSeek alone was reported to have produced around 150,000 interactions, while Moonshot AI and MiniMax generated millions more.
These claims have stirred debate within the industry, as distillation is commonly practiced. Notably, Elon Musk testified in federal court that his company xAI had "partially" utilized OpenAI models for training its own AI system, Grok.
In response to these issues, Congressman Bill Huizenga introduced a bill aimed at preventing foreign adversaries from extracting technical specifications from U.S. AI models, which includes export restrictions and sanctions against entities that unlawfully access AI products. Furthermore, Nvidia announced plans in December 2025 to develop technology to verify processor locations amid concerns over smuggling operations into China.
