Anthropic reveals industrial-scale Claude distillation attacks by seven Chinese AI labs
Anthropic has disclosed that seven China-based AI laboratories, including Alibaba, DeepSeek, and Moonshot, conducted large-scale illicit distillation campaigns against its Claude models. These operations, detected since February 2026, involved the unauthorized extraction of model capabilities, specifically chain-of-thought reasoning and agentic skills, to train rival systems. The attacks…
Key points
- Anthropic identified six illicit distillation campaigns by seven Chinese labs, including Alibaba, DeepSeek, and Moonshot, since February 2026.
- The largest attack, GTG-16005, involved 151 million exchanges and 3,500 fraudulent accounts targeting Claude Opus 4.6 and 4.7 reasoning.
- Labs used proxy services, stolen credentials, and silent traffic rerouting to harvest chain-of-thought data for unauthorized model training.
The most significant incident, labeled GTG-16005, involved Alibaba-affiliated operators harvesting 151 million exchanges between May and July 2026, peaking at 3 million daily requests. Other campaigns included Moonshot and DeepSeek silently relaying customer queries to Claude to capture training data, while Z.ai and Xiaomi employed similar replay and extraction pipelines. Anthropic noted that these labs also purchased user transcripts from third-party resellers who harvested data from proxy networks.
In response, Anthropic has implemented technical defenses, including updating Claude to summarize internal reasoning before responding to reduce the utility of stolen transcripts. The company also introduced "preserved thinking" in Fable 5.1 to prevent prompt manipulation. This disclosure aligns with recent accusations from U.S. intelligence agencies regarding systematic extraction of proprietary AI capabilities by Chinese firms.
The story so far
10 episodes →- Anthropic reveals industrial-scale Claude distillation attacks by seven Chinese AI labsthis story
China’s AI labs siphoned Claude’s intelligence?
yourstory.com · 11 September 2026
Loading the full article…
This text was published by yourstory.com and written by Team YS. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
9sources- Hacker News discussion · 6 pointsnews.ycombinator.com
- DeepSeek And Alibaba Are Closing The AI Gap. Anthropic Accuses them Of Using Claude To Help Train Their Models.Press · ibtimes.com ·
- Anthropic disrupts massive distillation attack from Chinese AI labs targeting ClaudePress · cryptobriefing.com ·
- Anthropic Says Seven China-Based AI Labs Ran Industrial-Scale Claude Distillation AttacksPress · thehackernews.com ·
- Anthropic opens the files on global Claude misuseNewsletter · The Rundown AI ·
- Chinese AI labs secretly used millions of Claude exchanges to train their models, Anthropic saysPress · CNBC Technology ·
- China's star AI labs routed user requests to Claude at least 35 million times in the summer: AnthropicPress · businessinsider.com ·
- Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeekPress · TechCrunch AI ·
- Anthropic caught Chinese AI labs carrying out massive ‘distillation attack’ to rip off its technologyPress · nypost.com ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Introducing Mistral Large 4 · 7 src
- OpenAI boosts GPT-6 speed by 50% in 28 Days update · 1 src
- Reflection AI debuts Beam, a 501B open-weight model rivaling GLM 5.2 with 3-4x less compute · 10 src
- Author tests four AI assistants on flawed forecasting scenarios · 1 src
- Anthropic prompts users to share voice data to enhance AI models · 8 src
Comments
via GitHub Discussions