Key facts
- Anthropic has accused Alibaba of conducting the largest known campaign to illicitly extract capabilities from its Claude AI model.
- The alleged campaign involved over 28.8 million exchanges using nearly 25,000 fraudulent accounts.
- Anthropic claims the attacks targeted valuable capabilities like agentic reasoning and software engineering.
- The AI firm recommended legislative actions including antitrust law updates, chip export controls, and penalties for Chinese labs.
- Alibaba's stock reportedly fell 3% following the accusations.
Anthropic has accused Chinese firm Alibaba of launching the largest known campaign to illicitly extract capabilities from its Claude AI model. The AI company shared evidence in a letter to Senators Tim Scott and Elizabeth Warren, detailing that between April 22 and June 5, operators affiliated with Alibaba and its AI lab allegedly used nearly 25,000 fraudulent accounts to generate over 28.8 million exchanges with Claude.
According to Anthropic, this campaign violated Claude's terms of service and access restrictions, targeting valuable capabilities such as agentic reasoning and software engineering. The firm stated that Alibaba employed obfuscation techniques and proxy networks to evade detection, aiming to replicate Claude's features without incurring the significant research and development costs.
Anthropic highlighted that these alleged attacks occurred after President Donald Trump had warned against such industrial-scale AI theft. The company also noted that OpenAI and Google have reported similar attacks on their models by Chinese firms. Anthropic suggested that these widespread distillation attacks effectively subsidize geopolitical competitors with hundreds of billions of dollars in American investment.
Alibaba, which is listed on the New York Stock Exchange, has reportedly denied any military affiliation and stated its business is focused on retail, logistics, and IT. However, Anthropic remains unconvinced and has urged Congress to pass legislation to deter such activities. Anthropic's recommendations include updating antitrust laws to facilitate information sharing among AI firms, imposing stricter export controls on advanced chips to China, and implementing penalties for Chinese labs engaging in such behavior.
Anthropic warned that if the US does not maintain its lead, China could develop advanced cyber capabilities and release AI models with weak safeguards that adversaries could exploit. Alibaba's Qwen family of AI models is noted as being popular in China, with over 700 million downloads. Alibaba's stock reportedly dropped 3% following Anthropic's accusations.