Anthropic: Alibaba, Moonshot, DeepSeek Ran Illicit Claude Distillation

Anthropic's threat report says Chinese AI labs ran illicit distillation campaigns, including over 151 million Claude exchanges tied to Alibaba, to train rival models.

@sharedot

Anthropic: Alibaba, Moonshot, DeepSeek Ran Illicit Claude Distillation

By @sharedot · · 7 pages

  • AI
  • Anthropic
  • Distillation
  • AI Safety

Anthropic's threat report says Chinese AI labs ran illicit distillation campaigns, including over 151 million Claude exchanges tied to Alibaba, to train rival models.

What Happened

Anthropic released a threat intelligence report on Thursday accusing China-based AI labs including Alibaba, Moonshot AI, and DeepSeek of "illicit distillation" — using Claude's outputs to train and improve their own models without authorization.

The company says it detected and disrupted the activity, which ran from December 2025 through August 2026.

According to the report, some exchanges included sensitive information from individual users, major multinational companies, and state-affiliated actors, practices Anthropic says are "likely inconsistent with privacy laws and the labs' own terms of service."

Why It Is Surprising

The scale and the alleged customer deception set this apart from routine terms-of-service disputes.

CNBC reports the Alibaba-linked campaign was the largest distillation effort Anthropic says it has ever measured, spanning more than 151 million exchanges between May and July and peaking at nearly 3 million exchanges per day from over 3,500 fraudulent accounts.

Moonshot, per the report, silently forwarded customer requests intended for its Kimi models to Claude, then displayed Claude's responses to users who believed they were talking to Kimi — relaying nearly 300,000 requests in a 10-day window.

The Evidence

The report attributes more than 23 million exchanges to Moonshot between May and July, with traffic routed through 5,380 accounts Anthropic deemed fraudulent, most apparently located in Singapore and Japan.

Anthropic says Moonshot saved exchanges and extracted Claude's reasoning transcripts as training data, and that it does not know whether Moonshot told customers their requests were being sent to Anthropic.

DeepSeek, according to the report, used similar tactics without notifying its own customers, with more than 12 million distillation attacks over 14 days in July 2026. The report also names several other major Chinese AI companies.

The Stakes

Anthropic's findings land as AI competition between US and Chinese labs intensifies and as data-handling practices face growing scrutiny. The report says Alibaba also used Claude for broader AI research, including reinforcement learning and model architecture, beyond straight distillation.

Because some forwarded customer prompts contained sensitive information from individuals, multinationals, and state-affiliated actors, the fallout could extend beyond model training into privacy-law exposure and customer trust.

None of the companies named — Alibaba, Moonshot, DeepSeek, Xiaomi, or Anthropic — responded immediately to CNBC's requests for comment.

What Comes Next

Anthropic says it has disrupted the activity, but the allegations are likely to fuel regulatory scrutiny, contractual responses, and new enforcement mechanisms around how frontier-model outputs are used.

How Alibaba, Moonshot, and DeepSeek respond, and whether regulators weigh in, could shape data-sharing rules across the industry.

Sources

  1. cnbc.com › Chinese AI labs secretly used millions of Claude exchanges to train their models, Anthropic says
  2. briefs.co › Anthropic Accuses Chinese AI Labs of Data Misuse
  3. blockchain.news › Anthropic: Chinese Labs Used Claude Data for Training

More on AI