Anthropic accuses Alibaba of the largest distillation campaign against Claude

What happened?

Anthropic, the artificial intelligence startup founded by former OpenAI members, has accused Alibaba of orchestrating the largest AI model distillation campaign against its Claude assistant. According to a letter sent to senators and senior White House officials, of which Bloomberg has obtained a copy, operators linked to Alibaba's Qwen lab used approximately 25,000 fraudulent accounts to extract Claude's capabilities between April and June 2025.

Distillation is a technique where a smaller model is trained using the outputs of a larger model, often without authorization. In this case, Anthropic claims the attackers used multiple accounts to send massive queries to Claude, collect its responses, and then use them to train competing models.

Why is this important?

This accusation is not an isolated incident but reflects an escalation in the trade and technology war between the United States and China. Unauthorized distillation of AI models has become common practice, but the scale of this case — 25,000 accounts — makes it the largest known to date. If confirmed, it would imply that Alibaba, one of China's tech giants, is using systematic methods to appropriate the intellectual property of US companies.

Moreover, the timing of the complaint is key: it comes amid Washington's export restrictions on AI chips to Beijing, which has led Chinese companies to seek alternatives to maintain competitiveness. Distillation of US models could be one such avenue.

What consequences will it have?

The consequences could be multiple. First, Anthropic is likely to pursue legal and regulatory measures against Alibaba, which could lead to trade sanctions or additional restrictions. The letter to the White House suggests Anthropic expects a government response, possibly in the form of new regulations that classify unauthorized distillation as a form of intellectual property theft.

Second, this case could harden the US stance toward Chinese AI, accelerating technological decoupling. Companies like Alibaba, Baidu, or Tencent could face even greater scrutiny in their global operations.

Finally, for the AI industry at large, this incident underscores the need for more robust protection mechanisms, such as early detection of anomalous queries and implementation of technical barriers against distillation.

What should readers know?

• Distillation is not illegal per se, but it is when done in violation of terms of service or intellectual property laws. Anthropic claims the fraudulent accounts violated its terms of use.
• This is not the first time Chinese companies have been accused of distillation. OpenAI and other startups have reported similar practices, though none of this magnitude.
• The geopolitical context is crucial: the US-China rivalry in AI is intensifying, and cases like this fuel the narrative that Beijing is willing to use questionable methods to achieve technological parity.
• Alibaba's response has not been made public yet, but it is expected to deny or downplay the accusations, arguing that these were independent actors not officially linked to Qwen.

“This is the largest distillation campaign we have seen against a US model. If no action is taken, it will set a dangerous precedent for intellectual property in the AI era,” sources close to Anthropic say.

Analysis

Anthropic's accusation highlights a structural problem in the AI ecosystem: the difficulty of protecting models when they are accessible through public APIs. While companies can implement rate limits and pattern detection, sophisticated attackers can bypass these defenses using distributed accounts and IP rotation.

From a business perspective, this case could lead to a rethink of business models based on open APIs. Companies like Anthropic, OpenAI, or Google could tighten access to their models, limiting free trials or requiring stricter identity verification.

On the regulatory front, the US administration could use this incident to justify new restrictions on the export of AI services, not just hardware. Cross-border model distillation could be considered a national security threat.