Anthropic accuses Alibaba of large-scale model distillation attack
Anthropic accused Alibaba of conducting an illegal model distillation attack using approximately 25,000 fraudulent accounts to extract capabilities from its Claude AI model through over 28.8 million interactions between April and June 2026. Anthropic called it the largest such attack against the company and urged coordinated government and industry action to combat distillation attacks.
What happened
Anthropic accused Alibaba of conducting an illegal model distillation attack on its Claude AI model. According to a letter dated June 10, 2026 sent to the U.S. Senate Committee on Banking, Housing and Urban Affairs, Alibaba used approximately 25,000 fraudulent accounts to conduct over 28.8 million interactions with Claude between April 22 and June 5, 2026.
Anthropic stated this was 'the largest-scale distillation attack known to have been conducted against Anthropic to date.' Model distillation involves using outputs from a larger AI model to train a smaller model, allowing developers to create powerful AI tools at lower cost.
Reported impact
The attack was described as aimed at illegally extracting capabilities from Anthropic's advanced 'Mythos Preview' features. Anthropic stated that such distillation attacks are conducted 'illegally, systematically and at industrial scale' to extract AI model capabilities from U.S. frontier labs and repackage them as proprietary technology without bearing the research and training costs.
News of the accusation caused Alibaba's Hong Kong stock price to fall 4.43% on June 25, 2026.
Index analysis
The incident followed Anthropic's earlier disclosure in February 2026 of three 'industrial-grade' distillation operations by Chinese AI developers DeepSeek, Moonshot AI, and MiniMax. Anthropic indicated the scale and complexity of such attacks were continuously escalating.
Anthropic, OpenAI, and Google had begun sharing information regarding model distillation attacks. Anthropic emphasized the importance of government and industry coordination in combating such attacks.
Governance controls
Anthropic stated that stopping such attacks required 'swift, coordinated action' from industry participants, policymakers and the global AI community. In April 2026, the White House had accused China of stealing AI intellectual property at industrial scale.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.