Anthropic, a US-based AI company, has accused Alibaba of conducting the largest known distillation attack on its Claude AI model to date. The alleged attack involved nearly 25,000 fake accounts and generated over 28.8 million exchanges between April and June 2026.

The Distillation Attack

Distillation attacks are a method used by AI companies to improperly enhance their models. They typically involve training a less capable AI on the outputs of a stronger one, allowing competitors to more quickly master advanced capabilities and corresponding technologies. In this case, Anthropic claims that operators linked to Alibaba's Qwen lab used nearly 25,000 fraudulent accounts to extract Claude's software engineering and agentic reasoning capabilities.

The scale of the operation is significant, with over 28.8 million interactions with Claude through nearly 25,000 fake accounts. This marks a substantial increase from previous allegations in February, which targeted smaller Chinese AI startups, including DeepSeek, MiniMax, and Moonshot AI. Those labs collectively generated more than 16 million exchanges through about 24,000 fake accounts.

Background and Context

Anthropic has been vocal about the threat of distillation attacks in recent months. In February, the company reported a separate case involving a Chinese state-sponsored group that used its Claude Code tool for outright cyber espionage. That earlier operation targeted nearly 30 different entities, with remarkably little human involvement.

The November case was about using Claude as a tool for espionage, while the February case was about stealing Claude itself or at least its capabilities. Both are violations, but they represent fundamentally different threat vectors that Anthropic now has to defend against simultaneously.

Why it Matters to the Industry

The distillation attack highlights growing concerns about AI security and IP protection in the industry. As AI models become increasingly powerful and valuable, companies like Alibaba are seeking ways to extract their capabilities without investing in research and development themselves. This raises urgent questions about how to protect AI technologies from malicious use and intellectual property leaks.

The consequences of such attacks can be significant, with unauthorised distillation costing Silicon Valley labs billions of dollars. The threat of cheaper imitation products from China that siphon away customers is a material risk for companies heading to public markets. In this case, Anthropic's valuation has reached $965 billion after a recent funding round, making it a prime target for competitors seeking to replicate its capabilities.

What Comes Next

The allegations against Alibaba have sparked a response from lawmakers in the US. Senators Bill Hagerty and Andy Kim plan to introduce an amendment to must-pass defence legislation that would blacklist or sanction any Chinese firm found to be improperly accessing US AI model output. A related bipartisan bill in the House is also being considered, though its prospects are uncertain.

Anthropic has called for government support in combating distillation attacks, urging the Trump administration to clarify antitrust guidelines so US labs can share more information about distillation attempts. The company has also reiterated its support for export controls on advanced AI chips and penalties against firms that use the technique.

Key Facts

  • Anthropic accused Alibaba of conducting the largest known distillation attack on its Claude AI model to date.
  • The alleged attack involved nearly 25,000 fake accounts and generated over 28.8 million exchanges between April and June 2026.
  • Distillation attacks involve training a less capable AI on the outputs of a stronger one to improperly enhance their models.
  • Anthropic has been vocal about the threat of distillation attacks in recent months, reporting separate cases involving Chinese state-sponsored groups.
  • The allegations against Alibaba have sparked a response from lawmakers in the US, with proposed legislation aimed at blacklisting or sanctioning Chinese firms found to be improperly accessing US AI model output.