Anthropic Accuses Alibaba of 28.8 Million-Query AI Distillation Attack
Updated
Updated · CNBC · Jun 24
Anthropic Accuses Alibaba of 28.8 Million-Query AI Distillation Attack
3 articles · Updated · CNBC · Jun 24
Summary
A June 10 letter to the Senate Banking Committee says Alibaba-affiliated operators made 28.8 million exchanges with Anthropic models between April 22 and June 5, using about 25,000 fraudulent accounts.
Anthropic called it the largest known distillation attack it has faced, alleging Alibaba’s Qwen-linked operators tried to extract Claude’s capabilities by training weaker models on outputs from stronger ones.
The complaint reaches Congress two months after the White House science office said it would help AI companies detect and coordinate against industrial-scale distillation; Anthropic said Alibaba ignored those warnings.
The filing also comes as Anthropic is separately negotiating with the Trump administration over an export-control order that suspended foreign nationals’ access to its latest Claude models, Fable 5 and Mythos 5.
As China masters AI 'distillation,' are America's most advanced digital secrets fundamentally indefensible?
Beyond code theft, how can the US protect its AI's core reasoning abilities from foreign replication?
The 2026 AI Distillation Showdown: U.S.-China Rivalry, Legal Ambiguity, and the Battle for Model Supremacy
Overview
In February 2026, OpenAI accused Chinese AI labs, including DeepSeek, of running large-scale distillation campaigns to extract capabilities from advanced U.S. AI models like ChatGPT. These Chinese companies used new, obfuscated methods to bypass restrictions and were described as 'free-riding' on U.S. technology. The White House responded with a memo naming China as a key actor in these campaigns, following a congressional report that detailed China’s use of both legal and illegal tactics to boost its AI and semiconductor industries. This series of events highlights growing tensions and the complex competition between the U.S. and China in AI development.