Anthropic, the San Francisco-based AI safety company, has formally accused China’s Alibaba Group of orchestrating a large-scale campaign to copy its Claude AI models through a technique known as model distillation.
In a letter sent June 10 to the U.S. Senate Committee on Banking, Housing, and Urban Affairs, Anthropic alleged that operators affiliated with Alibaba and its AI research division ran approximately 25,000 fraudulent accounts on the Claude platform between April 22 and June 5, 2026, generating 28.8 million exchanges with the model.
Anthopic described the campaign as “the largest known attack of its kind” ever carried out against its systems.
What Is Model Distillation?
Model distillation is a technique in which a less capable AI system is trained on outputs generated by a more powerful one, allowing it to replicate sophisticated capabilities at a fraction of the cost of developing them independently.
Anthopic alleged that Alibaba used this method to extract capabilities from its frontier Mythos Preview model — specifically targeting software engineering and agentic reasoning abilities — with the apparent goal of improving Alibaba’s own Qwen AI model.
According to Anthropic’s letter, a successful extraction could “help accelerate China’s ability to reach the capabilities” of its most advanced systems.
Alibaba Denies the Allegations
Alibaba did not respond to media requests for comment at the time of publication. The company has previously denied wrongdoing when pressed. The allegations have not been independently verified.
A New Category of Intellectual Property Risk
The case draws attention to a growing vulnerability in the AI industry. Unlike traditional data theft, distillation attacks require no access to a model’s source code or internal weights — only sustained, high-volume interaction with a publicly available API.
Analysts have described this as “a new category of AI-specific intellectual property vulnerability,” with implications for any company that exposes advanced AI capabilities through a public interface.
Why It Matters for Georgia
As of mid-2026, Georgia has no dedicated AI law, no regulatory authority, and no comprehensive national AI strategy. The government has committed USD 18.4 million over 2026–2029 to establishing an AI research and competence centre — a sign that the sector is becoming a policy priority.
The Anthropic-Alibaba dispute highlights a legal gap that Georgia, like most countries, has yet to address: existing intellectual property and cybercrime laws were not written with AI distillation in mind. For Georgian policymakers working toward EU regulatory alignment, this case presents a concrete argument for including explicit protections against unauthorized model extraction in any future AI legislation.