Anthropic Report Details Massive Chinese Efforts to Extract Claude Capabilities

Key points

• Anthropic identified large-scale unauthorized distillation efforts by Chinese AI companies, including Alibaba, Moonshot AI, and DeepSeek.
• The report details over 151 million exchanges with Claude by Alibaba between May and July, peaking at nearly 3 million daily interactions.
• Moonshot AI allegedly routed customer requests to Claude without disclosure, using the outputs to train its Kimi models.
• DeepSeek was found to have transferred over 12 million exchanges to Claude in a 14-day period in July 2026.
• Anthropic stated that some exchanges contained sensitive information from users, multinational corporations, and state-affiliated actors.

Background

This development follows earlier U.S. federal agency warnings in September 2026 about industrial-scale trade-secret theft by Chinese AI firms. Those reports accused companies like DeepSeek and Alibaba of systematically copying American models since 2024. Additionally, Anthropic has faced recent legal challenges regarding copyright infringement in its training data, highlighting ongoing scrutiny of its data practices.

Why it matters

The report highlights significant risks to data privacy and intellectual property in the global AI landscape. It underscores the competitive tensions between the U.S. and China, where unauthorized model distillation could lead to advanced capabilities being replicated without proper safeguards. This may influence future regulatory actions and corporate security measures in the AI sector.

What to watch

Anthropic has stated it disrupted these activities, but the long-term impact on model security remains unclear. Regulators may investigate the privacy violations mentioned in the report. The named companies have not yet responded to requests for comment, so official reactions are pending.