Anthropic, a leading AI company, has accused Alibaba of conducting a large-scale, illicit extraction of its AI model capabilities. The accusation, detailed in a letter to U.S. senators, alleges that Alibaba-affiliated operators carried out the largest known distillation attack on Anthropic to date.
Core Facts and Immediate Action
Anthropic's head of policy, Sarah Heck, wrote a letter to Sen. Tim Scott (R-S.C.) and Sen. Elizabeth Warren (D-Mass.) on June 10, accusing Alibaba of illicitly extracting capabilities from its Claude AI model. The letter, obtained by multiple outlets, states that between April 22 and June 5, Alibaba-affiliated operators conducted 28.8 million exchanges with Claude using nearly 25,000 fraudulent accounts. Anthropic described this as a "distillation attack," where advanced AI models are used to train less advanced ones without proper authorization.
Deeper Dive and Context
What is a Distillation Attack?
Distillation attacks involve using outputs from a more advanced AI model to train a less capable model, effectively bypassing the need for extensive research and development. Anthropic claims Alibaba's actions were systematic and industrial in scale, aiming to repurpose U.S. AI capabilities for its own models, such as Alibaba Cloud's Qwen LLMs.
Anthropic's Request for Legislation
In the letter, Heck called for legislation to prevent such attacks, including limiting China's access to advanced U.S. computing infrastructure and penalizing entities that conduct these attacks. This comes after the U.S. government imposed export controls on Anthropic's Fable 5 model, citing national security risks.
Alibaba's Response
As of the reporting, Alibaba has not publicly responded to the accusations. The company was recently added to a Pentagon blacklist, though the context of this addition is not directly related to the AI extraction claims.
Broader Implications
The accusation raises concerns about intellectual property theft and the competitive landscape in AI development. It also highlights ongoing tensions between U.S. and Chinese tech firms, particularly in the realm of artificial intelligence. Anthropic has emphasized the need for coordinated action between government and industry to combat such threats and maintain U.S. leadership in AI.