Anthropic Levels Serious Accusation at Alibaba: 28 Million Queries to Copy Claude
Originally published in Turkish. Read the original →
Anthropic has accused Chinese tech giant Alibaba of trying to distill the intelligence of Claude models through fake accounts. Following the developments, restrictions were placed on the Mythos and Fable models.
Allegations of Industrial Espionage in AI: The Model Distillation Crisis
As the race for global leadership in AI heats up, US-based AI startup Anthropic has made very serious accusations against Chinese tech giant Alibaba. According to Anthropic's official statement, Alibaba tried to illegally copy the capabilities of Claude, one of the company's most advanced AI models.
This allegation is being described as one of the largest and most organized "model distillation" attacks ever recorded in the AI world.
25,000 Fake Accounts and More Than 28 Million Queries
According to Anthropic, around 25,000 fake user accounts run by Alibaba's AI research group (the Qwen team) carried out a systematic operation between April 22 and June 5, 2026. Through these fake accounts, more than 28.8 million specially crafted queries were sent to Claude models.
What Is Model Distillation?
Model distillation is a method of systematically collecting the responses of a large, highly advanced AI model (the target model) and using that data to train a smaller, cheaper or self-developed model. Attackers use the target model's APIs almost like a "data mine," transferring the model's reasoning patterns into their own systems. This is one of the biggest gray areas in the AI world when it comes to protecting copyright and trade secrets.
The Main Target: The "Mythos Preview" Model
Anthropic officials stated that at the center of these coordinated attacks was the "Mythos Preview" model the company recently announced, which has revolutionary capabilities particularly in coding and digital security/cyber defense.
It is alleged that through this method Alibaba aimed to inject Mythos's cybersecurity and code development capabilities into its own open-source models. In its accusations, Anthropic also targeted not only Alibaba but the Chinese government as well, because of its strategic direction behind these activities.
Export Ban and Access Restriction on the Mythos and Fable Models
After the incident broke, the US government also acted quickly. Shortly after the letter was delivered, the US Department of Commerce, citing national security concerns, imposed very strict export restrictions on Anthropic's Mythos and Fable models.
In line with these restrictions, Anthropic had to temporarily suspend global general access to these advanced models (Mythos and Fable) until the legal and technical reviews with the federal government were settled.
Not the First Time China-Based Ventures Have Been Accused
This is not the first time Anthropic has faced attempts by Chinese rivals to copy its models. In February 2026, the company had also reported that three different Chinese AI firms, including DeepSeek, were collecting data with similar distillation methods.
This latest crisis shows once again that AI safety and intellectual property protection are no longer just a matter of commercial competition between companies but have turned into diplomatic and national security crises between countries.
